Remote sensing image segmentation method and device and related product

The remote sensing image segmentation method uses a deep learning network and attention mechanism to enhance feature extraction and probability distribution for efficient and accurate image segmentation across various tasks, addressing the cumbersome nature of existing technologies.

CN119091142APending Publication Date: 2024-12-06BEIJING SUPERMAP SOFTWARE CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411120529.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing remote sensing image processing technologies are cumbersome, requiring task-specific models, leading to increased research and development costs and time, and face issues like the salt and pepper phenomenon and poor small target detection.

Method used

A remote sensing image segmentation method using a deep learning network model for feature extraction, followed by detail enhancement with a pixel decoder and a decoder structure with an attention mechanism to generate a target parsing identifier, converting features into probability distributions for image segmentation.

Benefits of technology

This method reduces the need for task-specific models, enhances segmentation accuracy, and improves adaptability across different remote sensing tasks, reducing development costs and time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091142A_ABST
    Figure CN119091142A_ABST
Patent Text Reader

Abstract

The invention provides a remote sensing image segmentation method and apparatus, and a related product, which can be applied to the technical field of remote sensing, and comprises the steps of obtaining a to-be-segmented remote sensing image; performing feature extraction on the remote sensing image to be segmented through a deep learning network model to obtain a first feature extraction result, wherein the first feature extraction result comprises local features and regional features; performing detail enhancement on the first feature extraction result through a pixel decoder to obtain a second feature extraction result; generating a target analysis identifier by using a decoder structure based on an attention mechanism and the second feature extraction result, wherein the target analysis identifier is used for distinguishing different types of features; converting local features and regional features in the second feature extraction result into probability distribution based on the target analysis identifier; and performing image segmentation based on the probability distribution and the second feature extraction result to obtain an image segmentation result. Therefore, a model does not need to be independently designed for each task, and the image segmentation efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of remote sensing technology, and in particular to a remote sensing image segmentation method, device and related products. Background Art

[0002] With the continuous development of science and technology, remote sensing technology is also developing accordingly. Remote sensing images are increasingly widely used in many fields such as land cover classification, target detection, and change detection. Remote sensing images are a spatial form of repeated arrangement of individual targets. Remote sensing images can reflect the spatial distribution characteristics of objects. Many targets have certain repetitive relationships and form special combinations. It can be natural or artificially constructed. These features help to identify images, such as the different ridges of paddy fields and dry land in residential areas, and the neatly arranged tree crowns in orchards. When processing remote sensing images, related technologies will design specific models corresponding to different tasks. Related technologies are very cumbersome when processing remote sensing images, which in turn increases the cost and time of research and development.

[0003] Therefore, how to process remote sensing images more conveniently has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0004] Based on the above problems, the present application provides a remote sensing image segmentation method, device and related products to solve the very cumbersome problem of related technologies in processing remote sensing images.

[0005] In a first aspect, an embodiment of the present application provides a remote sensing image segmentation method, comprising:

[0006] Acquire a remote sensing image to be segmented;

[0007] Performing feature extraction on the remote sensing image to be segmented by a deep learning network model to obtain a first feature extraction result, wherein the first feature extraction result includes local features and regional features;

[0008] Performing detail enhancement on the first feature extraction result through a pixel decoder to obtain a second feature extraction result;

[0009] Generate a target parsing identifier using a decoder structure based on an attention mechanism and the second feature extraction result, wherein the target parsing identifier is used to distinguish features of different categories;

[0010] Converting the local features and regional features in the second feature extraction result into probability distribution based on the target parsing identifier;

[0011] Image segmentation is performed based on the probability distribution and the second feature extraction result to obtain an image segmentation result.

[0012] Optionally, before obtaining the remote sensing image to be segmented, the method further includes:

[0013] Acquire remote sensing images to be processed;

[0014] The remote sensing image to be processed is preprocessed to obtain the remote sensing image to be segmented.

[0015] Optionally, preprocessing the remote sensing image to be processed to obtain the remote sensing image to be segmented includes:

[0016] One or more of radiation correction, atmospheric correction, geometric correction and denoising is performed on the remote sensing image to be processed to obtain the remote sensing image to be segmented.

[0017] Optionally, the extracting features of the remote sensing image to be segmented by a deep learning network model to obtain a first feature extraction result includes:

[0018] The remote sensing image to be segmented is subjected to feature extraction through a deep learning network model to obtain one or more of terrain features, vegetation features, water body features and surface cover features.

[0019] Optionally, performing detail enhancement on the first feature extraction result by a pixel decoder to obtain a second feature extraction result includes:

[0020] The output results of different resolutions are fused through a convolutional network or an attention module to enhance the detail information and obtain the second feature extraction result.

[0021] Optionally, converting the local features and regional features in the second feature extraction result into probability distribution based on the target parsing identifier includes:

[0022] Based on the region attention mechanism, determining the key processing region from the second feature extraction result;

[0023] The local features and regional features of the key processing area are converted into probability distribution based on the target parsing identifier.

[0024] Optionally, the method further includes:

[0025] The deep learning network model is optimized.

[0026] In a second aspect, an embodiment of the present application provides a remote sensing image segmentation device, comprising:

[0027] An acquisition module, used for acquiring remote sensing images to be segmented;

[0028] A feature extraction module, used for performing feature extraction on the remote sensing image to be segmented through a deep learning network model to obtain a first feature extraction result, wherein the first feature extraction result includes local features and regional features;

[0029] A detail enhancement module, configured to perform detail enhancement on the first feature extraction result through a pixel decoder to obtain a second feature extraction result;

[0030] A parsing module, used to generate a target parsing identifier using a decoder structure based on an attention mechanism and the second feature extraction result, wherein the target parsing identifier is used to distinguish features of different categories;

[0031] A probability distribution module, used for converting the local features and regional features in the second feature extraction result into probability distribution based on the target parsing identifier;

[0032] A segmentation module is used to perform image segmentation based on the probability distribution and the second feature extraction result to obtain an image segmentation result.

[0033] In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, wherein:

[0034] The memory is used to store the computer program;

[0035] The processor is used to execute the computer program to implement the remote sensing image segmentation method as described above.

[0036] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the remote sensing image segmentation method as described above.

[0037] Compared with the prior art, this application has the following beneficial effects:

[0038] The present application first obtains a remote sensing image to be segmented, and extracts features from the remote sensing image to be segmented through a deep learning network model to obtain a first feature extraction result. The first feature extraction result includes local features and regional features. Then, the first feature extraction result is enhanced in detail through a pixel decoder to obtain a second feature extraction result, and a decoder structure based on an attention mechanism and the second feature extraction result are used to generate a target parsing identifier. The target parsing identifier is used to distinguish features of different categories. Finally, based on the target parsing identifier, the local features and regional features in the second feature extraction result are converted into a probability distribution, and image segmentation is performed based on the probability distribution and the second feature extraction result to obtain an image segmentation result. In this way, the method provided by the present application does not need to design a model for the task. When processing different image processing tasks, it only needs to operate according to the instructions of the target parsing identifier. The method provided by the present application can work across tasks, reduce the gap between different remote sensing image tasks, reduce the cost and time of research and development, and improve segmentation accuracy and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be 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 paying creative labor.

[0040] Figure 1 A flowchart of a remote sensing image segmentation method provided in an embodiment of the present application;

[0041] Figure 2 A remote sensing image segmentation flow chart based on a pixel decoder provided in an embodiment of the present application;

[0042] Figure 3 A schematic diagram of the structure of a remote sensing image segmentation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0043] As described above, when processing remote sensing images, relevant technologies will design specific models corresponding to different tasks. The relevant technologies are very cumbersome when processing remote sensing images, which in turn leads to increased research and development costs and time.

[0044] In addition, when processing remote sensing images, related technologies also include problems such as salt and pepper phenomenon, granularity phenomenon and poor effect on small targets. These problems make it difficult for the actual image processing results to meet the needs of practical applications.

[0045] After research, it was found that when processing remote sensing images, especially when segmenting remote sensing graphics, relevant technologies need to pre-design models corresponding to the segmentation tasks. For example, in the change monitoring task, a relevant model can be designed to detect the area of ​​change in surface cover; in land cover classification, a relevant model can be designed to identify and classify areas with specific features. The models used for different tasks are not universal, and in actual applications, the processing of remote sensing images often involves multiple tasks, which makes the processing of remote sensing images very cumbersome, which in turn increases the cost and time of research and development.

[0046] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0047] It is understandable that the method provided in the present application can be applied to a processing device, which is a processing device that can process an image, such as a terminal device or a server that can process an image. The method provided in the present application can be executed independently by a terminal device or a server, or can be applied to a network scenario in which a terminal device and a server communicate, and is executed by the cooperation of a terminal device and a server. Among them, the terminal device can be a computer, a mobile phone and other devices. The server can be understood as an application server or a Web server. In actual deployment, the server can be a stand-alone server or a cluster server.

[0048] Figure 1 A flow chart of a remote sensing image segmentation method provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0049] S101: Acquire a remote sensing image to be segmented.

[0050] In practical applications, the processing equipment needs to obtain remote sensing images (remote sensing images to be segmented) before using the backbone network to extract features from remote sensing images. Remote sensing images are a spatial form of repeated arrangement of individual targets, and remote sensing images can reflect the spatial distribution characteristics of objects. Many targets have a certain repetitive relationship, forming a special combination form. It can be natural or artificially constructed. These features help to identify images, such as the different ridges of paddy fields and dry land in residential areas, and the neatly arranged tree crowns in orchards.

[0051] In addition, since there are different ways to obtain the remote sensing image to be segmented, the embodiment of the present application can illustrate a possible acquisition method.

[0052] In one embodiment, before acquiring the remote sensing image to be segmented, the method further includes:

[0053] Acquire remote sensing images to be processed;

[0054] The remote sensing image to be processed is preprocessed to obtain the remote sensing image to be segmented.

[0055] In practical applications, remote sensing images are stored in memory in the form of data sets, and one remote sensing image corresponds to a set of remote sensing data. The processing device can obtain the original remote sensing image (remote sensing image to be processed) specified by the user and needs to be segmented from the memory. Then, the remote sensing data corresponding to the original remote sensing image is thoroughly preprocessed to ensure the quality of the image data, thereby obtaining the remote sensing image to be segmented.

[0056] In addition, since the image preprocessing methods are not the same, the embodiment of the present application can illustrate a possible processing method.

[0057] In one embodiment, preprocessing the remote sensing image to be processed to obtain the remote sensing image to be segmented includes:

[0058] One or more of radiation correction, atmospheric correction, geometric correction and denoising is performed on the remote sensing image to be processed to obtain the remote sensing image to be segmented.

[0059] In practical applications, the processing device preprocesses the remote sensing image to be processed to obtain the remote sensing image to be segmented. Preprocessing refers to the thorough processing of the data in advance. Preprocessing can include radiation correction, atmospheric correction, geometric correction, and denoising, etc., to ensure the quality of image data. In addition, the processing device can also perform image standardization processing on the remote sensing image to be processed, such as histogram equalization, to enhance the contrast of the image. In this way, the quality of the remote sensing image to be processed is improved through preprocessing, providing clear and accurate input for subsequent feature extraction.

[0060] S102: Performing feature extraction on the remote sensing image to be segmented through a deep learning network model to obtain a first feature extraction result, where the first feature extraction result includes local features and regional features.

[0061] In practical applications, key features can be extracted from the preprocessed remote sensing images to be segmented through a specially designed deep learning network structure. Specifically, the deep learning network model may include a backbone network, which may be, for example, ResNet or Swin Transformer. The processing device may input a data set I corresponding to a plurality of preprocessed remote sensing images to be segmented into the backbone network. The backbone network performs feature extraction on the preprocessed remote sensing images through a convolutional layer or an attention mechanism layer C, and obtains a first feature extraction result, which may specifically be a feature map F including local features and regional features. The calculation formula of the feature map F is shown in the following formula (1):

[0062] F=C(I)(1).

[0063] In addition, since the methods for obtaining the first feature extraction result are not the same, the embodiment of the present application can illustrate a possible acquisition method.

[0064] In one case, the step of extracting features of the remote sensing image to be segmented by a deep learning network model to obtain a first feature extraction result includes:

[0065] The remote sensing image to be segmented is subjected to feature extraction through a deep learning network model to obtain one or more of terrain features, vegetation features, water body features and surface cover features.

[0066] In practical applications, the processing device extracts features from the preprocessed remote sensing image through a deep learning network model to obtain a first feature extraction result. The feature extraction result includes small-scale local features (such as a single building, etc.) and large-scale regional features (large areas of land) in the remote sensing image to be processed, such as terrain features (such as mountains, plains), vegetation types (such as forests, grasslands), water bodies (such as rivers, lakes), and other surface covers (such as large buildings, roads).

[0067] S103: Perform detail enhancement on the first feature extraction result through a pixel decoder to obtain a second feature extraction result.

[0068] In practical applications, in order to enhance the details of features and make them have higher resolution and richer information, the embodiment of the present application inputs the feature map F output by the backbone network into the pixel decoder D, and uses a specific module to enhance the details of the extracted features.

[0069] In addition, since the methods for detail enhancement are not the same, the embodiment of the present application can illustrate a possible detail enhancement method.

[0070] In one case, performing detail enhancement on the first feature extraction result by a pixel decoder to obtain a second feature extraction result includes:

[0071] The output results of different resolutions are fused through a convolutional network or an attention module to enhance the detail information and obtain the second feature extraction result.

[0072] In practical applications, the processing device performs detail enhancement on the first feature extraction result through a specific module of the pixel decoder D, and obtains a second feature extraction result. The specific module may be a convolutional neural network or an attention mechanism module.

[0073] Specifically, the processing device may enhance the details of the feature extraction result by combining the upsampling operation with the feature fusion through the convolutional neural network or the attention mechanism module of the pixel decoder D, and obtain the second feature extraction result. The expression is shown in the following formula (2):

[0074] F′=D(F)

[0075]

[0076] Among them, F′ is the enhanced feature map with higher resolution and richer detail information. Upsample() represents the upsampling operation. Indicates the concatenation or fusion of feature maps. Scale fusion is implemented in the pixel decoder D, and multi-scale information is integrated through the feature pyramid P.

[0077] The above method can conduct a comprehensive analysis from local details to global features. This ensures that everything from small-scale local features (such as a single building) to large-scale regional features (large areas of land) can be effectively processed. The second feature extraction result obtained by detail enhancement has a higher resolution and can better reflect subtle features such as vegetation texture, building outlines, and small target information.

[0078] S104: Generate a target parsing identifier using a decoder structure based on an attention mechanism and the second feature extraction result, where the target parsing identifier is used to distinguish features of different categories;

[0079] The processing device generates a target parsing identifier by utilizing the Transformer decoder structure based on the attention mechanism and the second feature extraction result. The target parsing identifier is used to distinguish features of different categories. These target parsing identifiers can help accurately identify and distinguish different objects and phenomena in complex remote sensing images, such as distinguishing different types of vegetation or identifying the structure of a building. Specifically, the enhanced feature map F′ is input into the Transformer decoder T, which uses the self-attention mechanism and multi-head attention to parse the target. The calculation expression is shown in the following formula (3):

[0080] O=T(F′)(3)

[0081] Among them, O is the output of the decoder (target parsing identifier), which contains the parsing information of each target in the image and is used to distinguish the categories of local features and regional features.

[0082] In addition, since the methods for determining the key area are not the same, the embodiment of the present application can be described with respect to a possible determination method.

[0083] In one case, converting the local features and regional features in the second feature extraction result into a probability distribution based on the target parsing identifier may specifically include:

[0084] Based on the region attention mechanism, determining the key processing region from the second feature extraction result;

[0085] The local features and regional features of the key processing area are converted into probability distribution based on the target parsing identifier.

[0086] In practical applications, a regional attention mechanism can be introduced to focus on key areas in the image. For example, in the change detection task, we can focus on detecting the changed areas of the surface cover; in soil cover classification, we can identify and classify areas with specific characteristics, etc. In other words, the key areas are related to the detection requirements and can also be defined by the user. In this way, the embodiment of the present application can use the regional mask attention mechanism RMA in the Transformer decoder T to focus on the key areas in the image, and the calculation formula is shown in the following formula (4):

[0087]

[0088] Among them, Q, K, V are query, key and value matrices respectively, and M is the mask matrix generated according to the predicted mask, which is used to indicate the key areas that the model focuses on. The Softmax function normalizes the attention weights and converts them into probability distributions.

[0089] S105: Converting the local features and regional features in the second feature extraction result into probability distribution based on the target parsing identifier.

[0090] In practice, the output O of the Transformer decoder passes through a final classification layer C class To generate the category prediction of each pixel (local features and regional features in the second feature extraction result). Specifically, the calculation formula of the probability distribution is shown in the following formula (5):

[0091] Pc=C class (O) (5)

[0092] Among them, Pc is the category probability distribution of each pixel, which can be directly used for image segmentation.

[0093] S106: Perform image segmentation based on the probability distribution and the second feature extraction result to obtain an image segmentation result.

[0094] In practical applications, in combination with the above, the processing device can directly perform image segmentation based on the probability distribution and the second feature extraction result, and obtain the image segmentation result.

[0095] Another embodiment of the application can be used to illustrate a possible optimization method. In addition, since the optimization algorithm methods are different,

[0096] In one embodiment, the method further comprises:

[0097] The deep learning network model is optimized.

[0098] In practical applications, the deep learning network model can be optimized. Specifically, the algorithm is carefully optimized, including adjusting the processing order, improving the generation method of target parsing identifiers, and introducing a self-learning mechanism so that the algorithm can adapt to different remote sensing image features and changing conditions. In this way, the accuracy of segmentation and the generalization ability of the algorithm can be improved.

[0099] Figure 2 A flowchart of remote sensing image segmentation based on a pixel decoder is provided in an embodiment of the present application. Figure 2As shown in the figure, in the whole process, the model first extracts the key features of the remote sensing image through the backbone network, and then enhances the details of these features through the pixel decoder. Next, the decoder uses self-attention and region mask mechanisms to parse the targets in the image and focus on important areas. Finally, through scale fusion and classification layers, the model outputs the category prediction of each pixel to achieve accurate segmentation of the remote sensing image. This process involves a large number of matrix operations and nonlinear activation functions, which need to be trained and optimized through a deep learning framework. Through these detailed steps, this solution can effectively handle a variety of segmentation tasks in remote sensing images, providing an efficient and accurate image segmentation solution suitable for remote sensing applications such as land cover classification, target detection, and change monitoring.

[0100] In summary, the present application first obtains a remote sensing image to be segmented, and extracts features of the remote sensing image to be segmented through a deep learning network model to obtain a first feature extraction result. Among them, the first feature extraction result includes local features and regional features. Then, the first feature extraction result is enhanced in detail by a pixel decoder to obtain a second feature extraction result, and a decoder structure based on an attention mechanism and a second feature extraction result are used to generate a target parsing identifier. Among them, the target parsing identifier is used to distinguish features of different categories. Finally, based on the target parsing identifier, the local features and regional features in the second feature extraction result are converted into a probability distribution, and image segmentation is performed based on the probability distribution and the second feature extraction result to obtain an image segmentation result. In this way, the method provided by the present application does not need to design a model for the task. When processing different image processing tasks, it only needs to operate according to the instructions of the target parsing identifier. The method provided by the present application can work across tasks, reduce the gap between different remote sensing image tasks, reduce the cost and time of research and development, and improve segmentation accuracy and adaptability.

[0101] Based on the remote sensing image segmentation method provided in the above embodiment, the present application embodiment further provides a remote sensing image segmentation device. The remote sensing image segmentation device is described below in conjunction with the embodiments and drawings.

[0102] Figure 3 A schematic diagram of the structure of a remote sensing image segmentation device provided in an embodiment of the present application. Figure 3 As shown, the remote sensing image segmentation device 300 provided in the embodiment of the present application includes:

[0103] An acquisition module 301 is used to acquire a remote sensing image to be segmented;

[0104] A feature extraction module 302 is used to extract features of the remote sensing image to be segmented by using a deep learning network model to obtain a first feature extraction result, wherein the first feature extraction result includes local features and regional features;

[0105] A detail enhancement module 303, configured to perform detail enhancement on the first feature extraction result through a pixel decoder to obtain a second feature extraction result;

[0106] A parsing module 304, configured to generate a target parsing identifier using a decoder structure based on an attention mechanism and the second feature extraction result, wherein the target parsing identifier is used to distinguish features of different categories;

[0107] A probability distribution module 305, configured to convert the local features and regional features in the second feature extraction result into a probability distribution based on the target parsing identifier;

[0108] The segmentation module 306 is used to perform image segmentation based on the probability distribution and the second feature extraction result to obtain an image segmentation result.

[0109] As an implementation manner, with respect to how to obtain the remote sensing image to be segmented, the above-mentioned remote sensing image segmentation device 300 further includes: an acquisition submodule and a preprocessing module;

[0110] An acquisition submodule is used to acquire remote sensing images to be processed;

[0111] The preprocessing module is used to preprocess the remote sensing image to be processed to obtain the remote sensing image to be segmented.

[0112] As an implementation method, regarding how to perform image preprocessing, the preprocessing module is specifically used for:

[0113] One or more of radiation correction, atmospheric correction, geometric correction and denoising is performed on the remote sensing image to be processed to obtain the remote sensing image to be segmented.

[0114] As an implementation manner, regarding how to perform feature extraction, the feature extraction module 302 is specifically used for:

[0115] The remote sensing image to be segmented is subjected to feature extraction through a deep learning network model to obtain one or more of terrain features, vegetation features, water body features and surface cover features.

[0116] As an implementation manner, regarding how to perform detail enhancement, the detail enhancement module 303 is specifically used to:

[0117] The output results of different resolutions are fused through a convolutional network or an attention module to enhance the detail information and obtain the second feature extraction result.

[0118] As an implementation manner, regarding how to determine the probability distribution, the probability distribution module 305 is specifically used to:

[0119] Based on the region attention mechanism, determining the key processing region from the second feature extraction result;

[0120] The local features and regional features of the key processing area are converted into probability distribution based on the target parsing identifier.

[0121] As an implementation method, with respect to how to achieve algorithm optimization, the above-mentioned remote sensing image segmentation device 300 further includes: an algorithm optimization module;

[0122] The algorithm optimization module is used to optimize the deep learning network model.

[0123] In summary, the present application first obtains a remote sensing image to be segmented, and extracts features of the remote sensing image to be segmented through a deep learning network model to obtain a first feature extraction result. Among them, the first feature extraction result includes local features and regional features. Then, the first feature extraction result is enhanced in detail by a pixel decoder to obtain a second feature extraction result, and a decoder structure based on an attention mechanism and a second feature extraction result are used to generate a target parsing identifier. Among them, the target parsing identifier is used to distinguish features of different categories. Finally, based on the target parsing identifier, the local features and regional features in the second feature extraction result are converted into a probability distribution, and image segmentation is performed based on the probability distribution and the second feature extraction result to obtain an image segmentation result. In this way, the method provided by the present application does not need to design a model for the task. When processing different image processing tasks, it only needs to operate according to the instructions of the target parsing identifier. The method provided by the present application can work across tasks, reduce the gap between different remote sensing image tasks, reduce the cost and time of research and development, and improve segmentation accuracy and adaptability.

[0124] An embodiment of the present application also provides an electronic device, wherein the device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device performs the steps of the remote sensing image segmentation method described in any embodiment of the present application.

[0125] In practical applications, the computer-readable storage medium may be any combination of one or more computer-readable media, which may be a computer-readable signal medium or a computer-readable storage medium.

[0126] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, 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 thereof. In the present embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0127] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0128] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0129] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0130] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0131] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A remote sensing image segmentation method, characterized in that: include: Acquire a remote sensing image to be segmented; Performing feature extraction on the remote sensing image to be segmented through a deep learning network model to obtain a first feature extraction result, wherein the first feature extraction result includes local features and regional features; Performing detail enhancement on the first feature extraction result through a pixel decoder to obtain a second feature extraction result; Generate a target parsing identifier using a decoder structure based on an attention mechanism and the second feature extraction result, wherein the target parsing identifier is used to distinguish features of different categories; Converting the local features and regional features in the second feature extraction result into probability distribution based on the target parsing identifier; Image segmentation is performed based on the probability distribution and the second feature extraction result to obtain an image segmentation result.

2. The method according to claim 1, characterized in that Before obtaining the remote sensing image to be segmented, the method further includes: Acquire remote sensing images to be processed; The remote sensing image to be processed is preprocessed to obtain the remote sensing image to be segmented.

3. The method according to claim 2, characterized in that The preprocessing of the remote sensing image to be processed to obtain the remote sensing image to be segmented includes: One or more of radiation correction, atmospheric correction, geometric correction and denoising is performed on the remote sensing image to be processed to obtain the remote sensing image to be segmented.

4. The method according to claim 1, characterized in that: The step of extracting features of the remote sensing image to be segmented by a deep learning network model to obtain a first feature extraction result includes: The remote sensing image to be segmented is subjected to feature extraction through a deep learning network model to obtain one or more of terrain features, vegetation features, water body features and surface cover features.

5. The method according to claim 1, characterized in that The step of performing detail enhancement on the first feature extraction result by using a pixel decoder to obtain a second feature extraction result includes: The output results of different resolutions are fused through a convolutional network or an attention module to enhance the detail information and obtain the second feature extraction result.

6. The method according to claim 1, characterized in that The converting the local features and the regional features in the second feature extraction result into a probability distribution based on the target parsing identifier includes: Based on the region attention mechanism, determining the key processing region from the second feature extraction result; The local features and regional features of the key processing area are converted into probability distribution based on the target parsing identifier.

7. The method according to claim 1, characterized in that The method further comprises: The deep learning network model is optimized.

8. A remote sensing image segmentation device, characterized in that: include: An acquisition module, used for acquiring remote sensing images to be segmented; A feature extraction module, used for performing feature extraction on the remote sensing image to be segmented through a deep learning network model to obtain a first feature extraction result, wherein the first feature extraction result includes local features and regional features; A detail enhancement module, configured to perform detail enhancement on the first feature extraction result through a pixel decoder to obtain a second feature extraction result; A parsing module, used to generate a target parsing identifier using a decoder structure based on an attention mechanism and the second feature extraction result, wherein the target parsing identifier is used to distinguish features of different categories; A probability distribution module, used for converting the local features and regional features in the second feature extraction result into probability distribution based on the target parsing identifier; A segmentation module is used to perform image segmentation based on the probability distribution and the second feature extraction result to obtain an image segmentation result.

9. An electronic device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store the computer program; The processor is used to execute the computer program to implement the remote sensing image segmentation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein when the computer program is executed by a processor, the remote sensing image segmentation method according to any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Remote sensing image segmentation method and device, storage medium and server

    CN109801293A

  • Remote sensing image semantic segmentation method and system based on multi-scale information fusion

    CN113780296A

  • Multi-classification change detection method, device and equipment based on semantic change tag

    CN118262169A