Method and device for extracting sand dune shape based on remote sensing images

Abstract

The present disclosure provides a dune shape extraction method and device based on remote sensing images, the method comprising: obtaining a remote sensing image of a target dune; cropping the remote sensing image to a standard image set of a preset size; preprocessing the standard image set to generate a to-be-recognized image set; inputting the to-be-recognized image set into a pre-trained dune shape extraction model in a preset order, and outputting a corresponding recognition image set; overlapping cropping the recognition image set, and splicing them in the preset order to generate a recognition result image, the recognition result image including the edge shape of the target dune. In this way, the dune location information can be accurately grasped, and the dune shape can be accurately predicted.

Description

Technical Field

[0001] Embodiments of the present disclosure generally relate to the technical field of remote sensing image processing data preprocessing, and more specifically, to a method and apparatus for extracting sand dune shapes based on remote sensing images. Background Art

[0002] Desert areas are increasingly facing imminent threats and are vulnerable to the impacts of climate change, urbanization, land degradation, water shortages, etc. Deserts are extremely important habitats for specialized plant and animal communities. In addition, deserts have a wide range of minerals such as lithium, boron, etc. Understanding the changes in dune boundaries and their directions will also help in the ability to predict the movement of dunes. From a human perspective, if dunes are changing direction and moving towards densely populated areas, it will be a very dangerous natural behavior. Therefore, accurate monitoring of deserts is very necessary to develop and implement various effective measures to monitor dunes. However, most of the current dune extraction methods focus on manual outlining or using semi-automatic programs to extract dune contours and perform manual modifications, which requires a lot of manpower costs and cannot accurately extract dune semantic information over a large range of aspects.

[0003] When extracting dune contours, traditional edge extraction operators such as the Canny operator are not very targeted, and the extracted dune contour information has many noise points. Deep learning-based contour extraction algorithms such as HED and RCF can reduce noise points when extracting dune information, but they cannot accurately extract dune semantic information, so they also have disadvantages. In the context of lack of accurate semantic information extraction, although there is a method that uses the Unet network to extract dunes, the network is too simple, the number of ordinary convolutions in the upsampling and downsampling parts is large, and the feature map information is lost a lot. It is impossible to accurately grasp the location information of the dune feature map and cannot fully extract the accurate semantic information of the dunes in the desert. Considering that there are few dune samples in real scenes, how to use effective samples to train the network and make accurate predictions on large remote sensing images is particularly critical. Summary of the invention

[0004] According to an embodiment of the present disclosure, a dune shape extraction solution based on remote sensing images is provided, which can accurately grasp the location information of the dunes and accurately predict the shape of the dunes.

[0005] In a first aspect of the present disclosure, a method for extracting sand dune shape based on remote sensing images is provided, comprising:

[0006] Acquire remote sensing images of target sand dunes;

[0007] Cropping the remote sensing image to a standard image set of a preset size;

[0008] Preprocessing the standard image set to generate an image set to be identified;

[0009] Inputting the image set to be identified into a pre-trained dune shape extraction model in a preset order, and outputting a corresponding identification image set;

[0010] The recognition image set is overlapped and cropped, and spliced ​​according to the preset order to generate a recognition result image, which includes the edge shape of the target sand dune.

[0011] In some embodiments, the preprocessing of the standard image set to generate the image set to be identified includes:

[0012] The standard image set is filtered and denoised, and the grayscale value of a point in the standard image is replaced by the median of the neighborhood of the point.

[0013] In some embodiments, filtering and denoising the standard image set includes:

[0014] For remote sensing images, {x ij ,(i,j)∈I 2} represents the gray value of each point in the digital image, and the formula for image median filtering with a filter window of A is as follows:

[0015] y=Med{x ij}=Med{x (i+r),(j+s) (r,s)∈A(i,j)∈I 2}

[0016] Where Med{} is the median of the elements in the set, {x ij ,(i,j)∈I 2} represents the grayscale value of each pixel in the remote sensing image, I represents the size of the image neighborhood window, I is an odd number, A(i, j) represents the image area to be filtered in the image, and (r, s) represents the set of all pixels in the neighborhood window.

[0017] In some embodiments, the dune shape extraction model includes:

[0018] A deep residual dilation convolution encoding part and a layer-by-layer connection decoding part, wherein the deep residual dilation convolution encoding part includes five layers, and the output feature matrix dimension of each layer is added to the output feature matrix of the previous layer after being transposed convolution of the layer-by-layer connection decoding part layer by layer, forming a layer-by-layer connected network decoding structure.

[0019] In some embodiments, the deep residual expansion convolution coding part includes a five-layer structure, wherein the first layer structure includes three convolution layers with a convolution kernel size of 3*3, and a normalization layer and an activation layer arranged after the three convolution layers; the second layer structure differs from the first layer structure only in that the sizes of the convolution kernels are 1*1, 3*3, and 1*1, respectively, and the first layer structure and the second layer structure are connected through a maximum pooling layer; the third layer structure is the same as the second layer structure, and the feature map obtained by residual addition of the input feature map and the output feature map of the second layer structure is used as the input feature map of the third layer structure; the fourth layer structure includes a convolution layer with a convolution kernel size of 1*1 and an expansion convolution layer with expansion factors of 1, 2, and 5, respectively, and the fifth layer structure is the same as the fourth layer structure.

[0020] In some embodiments, the dune shape extraction model is trained in the following manner:

[0021] A preset number of images to be identified with pre-labeled dune edge shapes are used as training samples to train a pre-built dune shape extraction model, and a feature map with dune shapes is output as the recognition result;

[0022] Comparing the dune shape in the recognition result with the pre-marked dune edge shape, and in response to the error between the dune shape in the recognition result and the pre-marked dune edge shape being greater than a preset threshold, adjusting the parameters of each layer in the pre-constructed dune shape extraction model;

[0023] The above process is repeated until the error between the dune shape in the recognition result and the pre-marked dune edge shape is less than a preset threshold, and the training of the dune shape extraction model is completed.

[0024] In some embodiments, the step of cropping the remote sensing image to crop the remote sensing image into a standard image set of a preset size includes:

[0025] The remote sensing images are cropped with overlap so that the overlap ratio between adjacent standard images generated after cropping is 1-1 / r 2 , where r is the dilation rate in the dilated convolutional layer.

[0026] In a second aspect of the present disclosure, a device for extracting sand dune shape based on remote sensing images is provided, comprising:

[0027] An image acquisition module is used to acquire remote sensing images of target sand dunes;

[0028] An image cropping module crops the remote sensing image into a standard image set of a preset size;

[0029] An image preprocessing module, used for preprocessing the standard image set to generate an image set to be identified;

[0030] An image recognition module, used for inputting the image set to be recognized into a pre-trained dune shape extraction model in a preset order, and outputting a corresponding recognition image set;

[0031] The image stitching module is used to overlap and crop the recognition image set, and stitch them according to the preset order to generate a recognition result image, wherein the recognition result image includes the edge shape of the target sand dune.

[0032] In a third aspect of the present disclosure, an electronic device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, the method described above is implemented.

[0033] In a fourth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method described above is implemented.

[0034] Through the dune shape extraction method based on remote sensing images disclosed in the present invention, the location information of the dunes can be accurately grasped, and the shape of the dunes can be accurately predicted.

[0035] The contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0037] Figure 1 A flowchart of a method for extracting sand dune shapes based on remote sensing images according to an embodiment of the present disclosure is shown;

[0038] Figure 2 A schematic diagram showing the structure of a sand dune shape extraction device based on remote sensing images according to an embodiment of the present disclosure is shown;

[0039] Figure 3 A schematic diagram showing the structure of a sand dune shape extraction device based on remote sensing images according to an embodiment of the present disclosure is shown;

[0040] Figure 4 An example flow chart of a method for extracting sand dune shapes based on remote sensing images according to an embodiment of the present disclosure is shown;

[0041] Figure 5A schematic diagram showing the structure of a dune shape extraction model according to an embodiment of the present disclosure is shown;

[0042] Figure 6 A schematic diagram of image overlapping and cropping according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0044] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0045] The dune shape extraction method based on remote sensing images in the disclosed embodiment can accurately grasp the location information of the dunes and accurately predict the shape of the dunes.

[0046] Specifically, Figure 1 , which is a flow chart of a method for extracting sand dune shapes based on remote sensing images according to an embodiment of the present disclosure. Figure 4 An example flow chart of a method for extracting sand dune shapes based on remote sensing images according to an embodiment of the present disclosure is shown. Figure 1 and Figure 4 , the dune shape extraction method based on remote sensing images can include the following steps:

[0047] S101: Acquire a remote sensing image of a target sand dune.

[0048] The dune shape extraction method based on remote sensing images of the embodiment of the present disclosure can be applied to the dune shape extraction in desert areas, such as the shape extraction of crescent-shaped dunes. By extracting the shape of the dunes, the movement of the dunes can be predicted, thereby realizing the monitoring of the dunes. Of course, the method of the embodiment of the present disclosure can also be applied to the shape extraction of other terrains, such as rivers, lakes, etc.

[0049] In this embodiment, when the method disclosed in the present invention is used to extract the shape of a sand dune, it is first necessary to obtain a remote sensing image of the target sand dune. Preferably, the remote sensing image is a high-resolution remote sensing image. The remote sensing image includes an image of the target sand dune.

[0050] S102: Cropping the remote sensing image to a standard image set of a preset size.

[0051] In the model prediction process, if a large remote sensing image to be classified is directly input into the network model, it will cause memory overflow. Therefore, the image to be classified is generally cropped into a series of smaller images and input into the network for prediction. Then the prediction results are spliced ​​into a final result image in the order of cropping. Corresponding to the layer-by-layer connected deep residual dilated convolutional network model, the method of the embodiment of the present disclosure adopts the prediction of ignoring edges, that is, overlapping images are cropped and the strategy of ignoring edges is adopted when splicing.

[0052] In this embodiment, after acquiring the remote sensing image of the target sand dune, the acquired remote sensing image can be further cropped to crop the remote sensing image into a standard image set of preset size. Since the method of the embodiment of the present disclosure utilizes a neural network model to extract the shape of the sand dune, it is necessary to crop the remote sensing image into a standard image suitable for the neural network input, and in order to fully utilize the feature map information of each layer of the neural network model, give the dune remote sensing image a larger receptive field, accurately grasp the dune location information when extracting the sand dune information and classify it to obtain accurate prediction results, the neural network model of the embodiment of the present disclosure adopts a layer-by-layer connected deep residual hole convolution network. In order to adapt to the layer-by-layer connected deep residual hole convolution network, the remote sensing image needs to be cropped to crop the remote sensing image into a standard image set of preset size.

[0053] like Figure 6 FIG. 1 is a schematic diagram of image overlapping cropping according to an embodiment of the present disclosure. In this embodiment, when cropping a remote sensing image, overlapping cropping is required. Figure 6 The dotted part in represents the image size after overlapping cropping. Figure 6 The solid line part in represents the size of the useful information image.

[0054] like Figure 5 As shown, Figure 5 The structural schematic diagram of the dune shape extraction model of the embodiment of the present disclosure is shown. The dune shape extraction model of the present embodiment includes a deep residual dilation convolutional encoding part and a layer-by-layer connection decoding part. The deep residual dilation convolutional encoding part includes five layers. The output feature matrix dimension of each layer is added to the output feature matrix of the previous layer after being transposed convolution of the layer-by-layer connection decoding part layer by layer, forming a layer-by-layer connected network decoding structure.

[0055] The deep residual expansion convolution coding part includes a five-layer structure, wherein the first layer structure includes three convolution layers with a convolution kernel size of 3*3, and a normalization layer and an activation layer arranged after the three convolution layers; the second layer structure is different from the first layer structure only in that the sizes of the convolution kernels are 1*1, 3*3, and 1*1, respectively, and the first layer structure and the second layer structure are connected through a maximum pooling layer; the third layer structure is the same as the second layer structure, and the feature map obtained by residual addition of the input feature map and the output feature map of the second layer structure is used as the input feature map of the third layer structure; the fourth layer structure includes a convolution layer with a convolution kernel size of 1*1 and an expansion convolution layer with expansion factors of 1, 2, and 5, respectively, and the fifth layer structure is the same as the fourth layer structure.

[0056] The dilated convolution is defined as follows, where f(i) is the input information, g(i) is the output signal, h(i) represents a filter of length L, and r corresponds to the dilation rate, where r = 1 in standard convolution:

[0057]

[0058] Furthermore, the dune shape extraction model is trained in the following way:

[0059] A preset number of images to be identified with pre-labeled dune edge shapes are used as training samples to train a pre-built dune shape extraction model, and a feature map with dune shapes is output as the recognition result;

[0060] Comparing the dune shape in the recognition result with the pre-marked dune edge shape, and in response to the error between the dune shape in the recognition result and the pre-marked dune edge shape being greater than a preset threshold, adjusting the parameters of each layer in the pre-constructed dune shape extraction model;

[0061] The above process is repeated until the error between the dune shape in the recognition result and the pre-marked dune edge shape is less than a preset threshold, and the training of the dune shape extraction model is completed.

[0062] S103: Preprocess the standard image set to generate a set of images to be identified.

[0063] After the remote sensing image is cropped to generate a standard image set, the standard images in the standard image set are preprocessed to generate an image set to be identified. The preprocessing of the standard image set includes: filtering and denoising the standard image set, and replacing the gray value of a point in the standard image with the median of the neighborhood of the point.

[0064] Specifically, for remote sensing images, {x ij ,(i,j)∈I 2} represents the gray value of each point in the digital image, and the formula for image median filtering with a filter window of A is as follows:

[0065] y=Med{x ij}=Med{x (i+r),(j+s) (r,s)∈A(i,j)∈I 2}

[0066] Where Med{} is the median of the elements in the set, {x ij ,(i,j)∈I 2} represents the grayscale value of each pixel in the remote sensing image, I represents the size of the image neighborhood window, I is an odd number, A(i, j) represents the image area to be filtered in the image, and (r, s) represents the set of all pixels in the neighborhood window.

[0067] The cropped image is subjected to median filtering. The dune remote sensing image is disturbed by irregular gravel, the sand ridge line is discontinuous, and the dune slope forms point or patch noise. The median filtering algorithm is used for denoising. The median filtering algorithm can eliminate noise and protect the edge of the image, and has an image restoration effect. The median filtering can remove isolated noise points (such as gravel, etc.) and maintain the edge feature shape.

[0068] S104: Inputting the to-be-recognized image set into a pre-trained dune shape extraction model in a preset order, and outputting a corresponding recognition image set.

[0069] In this embodiment, after the image set to be identified is generated, the image set to be identified can be input into the pre-trained dune shape extraction model in a preset order, and the corresponding identification image set can be output. Among them, the images to be identified in the image set to be identified can be numbered according to the order of cutting, so as to determine the order of input into the dune shape extraction model. Alternatively, the order of input into the pre-trained dune shape extraction model can also be determined according to other methods. In this embodiment, the order of input into the pre-trained dune shape extraction model is determined in order to determine the order of the output identification images, and then determine the position of the order of the identification images in the final identification result image, so as to facilitate splicing.

[0070] S105: Overlapping and cropping the recognition image set, and splicing them according to the preset order to generate a recognition result image, wherein the recognition result image includes the edge shape of the target sand dune.

[0071] In this embodiment, after the images to be identified in the image set to be identified are identified and the identified image set is generated, since the identified images correspond to the images to be identified, the images to be identified are cropped in an overlapping manner, and therefore, the identified images also include additional information, such as Figure 6As shown, the recognition image is a square with a side length of A, and the useful information is a square with a side length of a, then A 2 -a 2 In addition, in this embodiment, when cropping a remote sensing image for recognition or cropping a recognition image to produce a recognition result image, the remote sensing image is cropped with overlap so that the overlap ratio between adjacent standard images generated after cropping is 1-1 / r. 2 , where r is the dilation rate in the dilated convolutional layer.

[0072] The dune shape extraction method based on remote sensing images in the disclosed embodiment can accurately grasp the location information of the dunes and accurately predict the shape of the dunes.

[0073] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present disclosure is not limited by the order of the actions described, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present disclosure.

[0074] The above is an introduction to the method embodiment. The following is a further explanation of the scheme disclosed in the present invention through an apparatus embodiment.

[0075] like Figure 2 FIG. 1 is a schematic diagram of the structure of a sand dune shape extraction device based on remote sensing images according to an embodiment of the present disclosure. The sand dune shape extraction device based on remote sensing images according to this embodiment includes:

[0076] An image acquisition module 201 is used to acquire a remote sensing image of a target sand dune;

[0077] An image cropping module 202 crops the remote sensing image into a standard image set of a preset size;

[0078] An image preprocessing module 203 is used to preprocess the standard image set to generate an image set to be identified;

[0079] An image recognition module 204 is used to input the image set to be recognized into a pre-trained dune shape extraction model in a preset order, and output a corresponding recognition image set;

[0080] The image stitching module 205 is used to overlap and crop the recognition image set, and stitch them according to the preset order to generate a recognition result image, wherein the recognition result image includes the edge shape of the target sand dune.

[0081] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0082] Figure 3 A schematic block diagram of an electronic device 300 that can be used to implement an embodiment of the present disclosure is shown. As shown, the device 300 includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 302 or loaded from a storage unit 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the device 300 can also be stored. The CPU 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0083] A number of components in the device 300 are connected to the I / O interface 305, including: an input unit 306, such as a keyboard, a mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a disk, an optical disk, etc.; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 309 allows the device 300 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0084] The processing unit 301 performs the various methods and processes described above, which are tangibly contained in a machine-readable medium, such as a storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 703 and executed by the CPU 301, one or more steps of the method described above can be performed. Alternatively, in other embodiments, the CPU 301 can be configured to perform the above method in any other suitable manner (e.g., by means of firmware).

[0085] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip systems (SOCs), load programmable logic devices (CPLDs), and the like.

[0086] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0087] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0088] In addition, although each operation is described in a specific order, this should be understood as requiring such operation to be performed in the specific order shown or in a sequential order, or requiring that all illustrated operations should be performed to obtain desired results. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single implementation in combination. On the contrary, the various features described in the context of a single implementation can also be implemented in multiple implementations individually or in any suitable sub-combination mode.

[0089] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A dune shape extraction method based on remote sensing images, characterized in that: include: Acquire remote sensing images of target sand dunes; Cropping the remote sensing image to a standard image set of a preset size; Preprocessing the standard image set to generate an image set to be identified; Inputting the to-be-recognized image set into a pre-trained dune shape extraction model in a preset order, and outputting a corresponding recognition image set; The dune shape extraction model comprises: a deep residual expansion convolution encoding part and a layer-by-layer connection decoding part, wherein the deep residual expansion convolution encoding part comprises five layers, and the output feature matrix dimension of each layer is added to the output feature matrix of the previous layer after being transposed convolution of the layer-by-layer connection decoding part layer by layer, so as to form a layer-by-layer connected network decoding structure; the deep residual expansion convolution encoding part comprises a five-layer structure, wherein the first layer structure comprises three convolution layers with a convolution kernel size of 3*3, and a normalization layer and an activation layer arranged after the three convolution layers; the second layer structure differs from the first layer structure only in that the sizes of the convolution kernels are 1*1, 3*3, and 1*1 respectively, and the first layer structure and the second layer structure are connected through a maximum pooling layer; the third layer structure is the same as the second layer structure, and the feature map obtained by residual addition of the input feature map and the output feature map of the second layer structure is used as the input feature map of the third layer structure; the fourth layer structure comprises a convolution layer with a convolution kernel size of 1*1 and an expansion convolution layer with expansion factors of 1, 2, and 5 respectively, and the fifth layer structure is the same as the fourth layer structure; The recognition image set is overlapped and cropped, and spliced ​​according to the preset order to generate a recognition result image, which includes the edge shape of the target sand dune.

2. The dune shape extraction method according to claim 1, characterized in that: The preprocessing of the standard image set to generate the image set to be identified includes: The standard image set is filtered and denoised, and the grayscale value of a point in the standard image is replaced by the median of the neighborhood of the point.

3. The dune shape extraction method according to claim 2, characterized in that: The filtering and denoising of the standard image set comprises: For remote sensing images, Representing the gray value of each point in the digital image, the formula for image median filtering with a filter window of A is as follows: Among them, Med{} is the median of the elements in the set. Represents the grayscale value of each pixel in the remote sensing image, I represents the size of the image neighborhood window, I is an odd number, A(i, j) represents the image area to be filtered in the image, and (r, s) represents the set of all pixels in the neighborhood window.

4. The dune shape extraction method according to claim 1, characterized in that: The dune shape extraction model is trained in the following way: A preset number of images to be identified with pre-labeled dune edge shapes are used as training samples to train the pre-built dune shape extraction model, and a feature map with dune shapes is output as the recognition result; Comparing the dune shape in the recognition result with the pre-marked dune edge shape, and in response to the error between the dune shape in the recognition result and the pre-marked dune edge shape being greater than a preset threshold, adjusting the parameters of each layer in the pre-constructed dune shape extraction model; The above process is repeated until the error between the dune shape in the recognition result and the pre-marked dune edge shape is less than a preset threshold, and the training of the dune shape extraction model is completed.

5. The dune shape extraction method according to claim 4, characterized in that: The step of cropping the remote sensing image to crop the remote sensing image into a standard image set of a preset size includes: The remote sensing images are cropped with overlap, so that the overlap ratio between adjacent standard images generated after cropping is , where r is the dilation rate in the dilated convolutional layer.

6. A device for extracting sand dune shape based on remote sensing images, characterized in that: include: An image acquisition module is used to acquire remote sensing images of target sand dunes; An image cropping module crops the remote sensing image into a standard image set of a preset size; An image preprocessing module, used for preprocessing the standard image set to generate an image set to be identified; An image recognition module, used for inputting the image set to be recognized into a pre-trained dune shape extraction model in a preset order, and outputting a corresponding recognition image set; The dune shape extraction model comprises: a deep residual expansion convolution encoding part and a layer-by-layer connection decoding part, wherein the deep residual expansion convolution encoding part comprises five layers, and the output feature matrix dimension of each layer is added to the output feature matrix of the previous layer after being transposed convolution of the layer-by-layer connection decoding part layer by layer, so as to form a layer-by-layer connected network decoding structure; the deep residual expansion convolution encoding part comprises a five-layer structure, wherein the first layer structure comprises three convolution layers with a convolution kernel size of 3*3, and a normalization layer and an activation layer arranged after the three convolution layers; the second layer structure differs from the first layer structure only in that the sizes of the convolution kernels are 1*1, 3*3, and 1*1 respectively, and the first layer structure and the second layer structure are connected through a maximum pooling layer; the third layer structure is the same as the second layer structure, and the feature map obtained by residual addition of the input feature map and the output feature map of the second layer structure is used as the input feature map of the third layer structure; the fourth layer structure comprises a convolution layer with a convolution kernel size of 1*1 and an expansion convolution layer with expansion factors of 1, 2, and 5 respectively, and the fifth layer structure is the same as the fourth layer structure; The image stitching module is used to overlap and crop the recognition image set, and stitch them according to the preset order to generate a recognition result image, wherein the recognition result image includes the edge shape of the target sand dune.

7. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images