Building contour marking method and device, and building contour segmentation method and device

Abstract

The invention discloses a building contour labeling method and device, and a building contour segmentation method and device, and the labeling method comprises the steps: obtaining a plurality of to-be-labeled images, so as to obtain an original data set; performing manual annotation on a part of the data set in the original data set to obtain a real annotation; processing the part of the data set by adopting a first image enhancement method to obtain first enhanced data, processing the part of the data set by adopting a second image enhancement method to obtain second enhanced data, and taking the first enhanced data, the second enhanced data and the part of the data set as an enhanced data set; inputting the enhanced data set into a pre-training large model to obtain a corresponding first prediction label; training the conversion network by adopting the first prediction label and the real label to obtain a trained conversion network; labeling the other part of the data set in the original data set by adopting the pre-trained large model and the trained conversion network to obtain a labeling result; and therefore, the manual labeling cost is greatly reduced.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a building outline marking method, a building outline segmentation method, a building outline marking device, a building outline segmentation device and a computer-readable storage medium. Background Art

[0002] In recent years, with the rapid development of remote sensing satellite technology, remote sensing images have shown extensive application potential in many industries, especially in the field of intelligent construction, where the prospects are even broader. In urban construction, accurate distribution information of buildings is crucial for planning and design, social demographics, and regional development analysis. By performing semantic segmentation on remote sensing images, we can efficiently and accurately extract urban composition and building distribution data, providing reliable technical support for urban planning, infrastructure layout, and engineering construction.

[0003] Traditional remote sensing building semantic segmentation methods mostly rely on manually designed feature algorithms based on image processing and machine learning, such as threshold, region or edge segmentation methods. These methods may be effective in specific scenarios, but they have poor applicability and rely heavily on human intervention. With the introduction of deep neural networks, remote sensing building segmentation technology has made significant progress in efficiency and accuracy. However, deep learning methods usually require a large amount of labeled data. Especially when processing high-resolution remote sensing images, pixel-level annotation is time-consuming and labor-intensive, greatly increasing labor costs. Summary of the invention

[0004] The present application aims to solve at least one of the technical problems in the above-mentioned technology to a certain extent. To this end, one purpose of the present application is to propose a building outline annotation method, which combines manual annotation and large model annotation, realizes the mapping between the two with the help of a conversion network, so as to generate high-quality annotation data on a large scale, thereby significantly reducing the cost of manual annotation.

[0005] The second objective of this application is to propose a building contour segmentation method.

[0006] The third objective of the present application is to provide a building outline marking device.

[0007] The fourth objective of the present application is to provide a building outline marking and segmentation device.

[0008] A fifth objective of the present application is to provide a computer-readable storage medium.

[0009] To achieve the above-mentioned objectives, the first aspect of the present application proposes a method for labeling building contours, comprising the following steps: acquiring multiple images to be labeled to obtain an original data set; manually labeling a portion of the original data set to obtain real labeled data; using a first image enhancement method to process the portion of the data set to obtain first enhanced data, and using a second image enhancement method to process the portion of the data set to obtain second enhanced data, and using the first enhanced data, the second enhanced data and a portion of the data set as an enhanced data set; inputting the enhanced data set into a pre-trained large model to obtain corresponding first predicted labeled data; using the first predicted labeled data and the real labeled data to train a conversion network to obtain a trained conversion network; using the pre-trained large model and the trained conversion network to label another portion of the original data set to obtain a labeling result.

[0010] According to the building outline labeling method of the embodiment of the present application, by combining manual labeling and large model labeling, the mapping between the two is achieved with the help of a conversion network to generate high-quality labeling data on a large scale, thereby significantly reducing the cost of manual labeling.

[0011] In addition, the building outline marking method proposed in the above embodiment of the present application may also have the following additional technical features:

[0012] Optionally, acquiring multiple images to be annotated to obtain an original data set includes: acquiring multiple remote sensing building images; and processing, cropping and filtering the multiple remote sensing building images to obtain an original data set.

[0013] Optionally, the first image enhancement method is random noise enhancement, and the second image enhancement method is random contrast enhancement.

[0014] Optionally, the conversion network is trained using the first predicted labeled data and the real labeled data to obtain a trained conversion network, including: inputting the first predicted labeled data into a generator to obtain second predicted labeled data; obtaining a first loss function based on the real labeled data and the second predicted labeled data, and obtaining a total loss function of the conversion network based on the first loss function and a second loss function corresponding to the discriminator; and training the generator using the total loss function to obtain a trained conversion network.

[0015] Optionally, the total loss function of the conversion network is obtained according to the following formula:

[0016] L=L1+λL GAN

[0017]

[0018] Among them, L represents the total loss function, λ represents the weight parameter of adaptive learning, represents the gradient of the loss function at the last layer of the generator, δ = 10 -6 represents the adjustment factor of the learning rate, x represents the real labeled data, represents the second predicted labeled data, and D represents the discriminator.

[0019] Optionally, the building outline labeling method further includes: inputting the labeling result into the discriminator so as to score each labeling data in the labeling result through the discriminator; and filtering out the labeling data with a score lower than a preset threshold to obtain final labeling data.

[0020] To achieve the above-mentioned purpose, the second aspect of the present application proposes a method for building contour segmentation, comprising the following steps: adopting the labeling method as described in any one of the first aspects to obtain the labeling result and its corresponding original image; inputting the labeling result and its corresponding original image into the segmentation model to be trained for training to obtain a trained segmentation model; obtaining the target image to be segmented; inputting the target image to be segmented into the trained segmentation module to obtain the corresponding segmentation result.

[0021] According to the building outline segmentation method of the embodiment of the present application, the labeling result is obtained by utilizing the above-mentioned building outline labeling method, thereby greatly reducing the cost of manual labeling and improving the accuracy and efficiency of segmentation.

[0022] To achieve the above-mentioned purpose, the third aspect of the present application proposes a building outline annotation device, including a first acquisition module, used to acquire multiple images to be annotated to obtain an original data set; a manual annotation module, used to manually annotate a part of the original data set to obtain real annotation data; a data processing module, used to use a first image enhancement method to process the part of the data set to obtain first enhanced data, and use a second image enhancement method to process the part of the data set to obtain second enhanced data, and use the first enhanced data, the second enhanced data and a part of the data set as an enhanced data set; a prediction annotation module, used to input the enhanced data set into a pre-trained large model to obtain corresponding first prediction annotation data; a conversion network training module, used to train a conversion network using the first prediction annotation data and the real annotation data to obtain a trained conversion network; an optimization annotation module, used to use the pre-trained large model and the trained conversion network to annotate another part of the original data set to obtain an annotation result.

[0023] To achieve the above-mentioned objectives, the fourth aspect of the present application proposes a building contour segmentation device, including: a second acquisition module, used to adopt the above-mentioned labeling method to obtain the labeling result and the corresponding original image; a segmentation model training module, used to input the labeling result and the corresponding original image into the segmentation model to be trained for training to obtain a trained segmentation model; a third acquisition module, used to acquire the target image to be segmented; a segmentation module, used to input the target image to be segmented into the trained segmentation module to obtain the corresponding segmentation result.

[0024] To achieve the above objectives, the fifth aspect of the present application proposes a computer-readable storage medium, on which a building outline labeling program is stored. When the building outline labeling program is executed, it implements the building outline labeling method as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A schematic diagram of a flow chart of a method for marking building outlines according to an embodiment of the present application;

[0026] Figure 2 A schematic diagram of a network structure of a building outline marking method according to an embodiment of the present application;

[0027] Figure 3 A schematic diagram of a flow chart of a building outline segmentation method according to an embodiment of the present application;

[0028] Figure 4 is a block diagram of a building outline marking device according to an embodiment of the present application;

[0029] Figure 5 4 is a block diagram of a building outline segmentation device according to an embodiment of the present application. DETAILED DESCRIPTION

[0030] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0031] In recent years, with the rise of large models, such as SAM (Segment Anything Model), its powerful feature extraction capabilities have performed well in dealing with zero-sample and small-sample problems; remote sensing technology in construction has begun to explore the secondary development of large models, aiming to reduce the cost of manual labeling while ensuring improved segmentation accuracy; the research goal in this field is to develop remote sensing building segmentation methods suitable for actual construction scenarios based on large models, taking into account both efficiency and accuracy, and providing innovative technical solutions for future intelligent construction management, dynamic planning and building life cycle management, accelerating the transformation and upgrading of the construction industry to intelligent construction.

[0032] To this end, the building outline annotation method proposed in the present application obtains a plurality of images to be annotated to obtain an original data set; manually annotates a portion of the original data set to obtain real annotation data; uses a first image enhancement method to process a portion of the data set to obtain first enhanced data, and uses a second image enhancement method to process a portion of the data set to obtain second enhanced data, and uses the first enhanced data, the second enhanced data and a portion of the data set as an enhanced data set; inputs the enhanced data set into a pre-trained large model to obtain corresponding first predicted annotation data; uses the first predicted annotation data and the real annotation data to train a conversion network to obtain a trained conversion network; uses the pre-trained large model and the trained conversion network to annotate another portion of the data set in the original data set to obtain an annotation result; thus, by combining manual annotation and large model annotation, the mapping between the two is achieved with the help of a conversion network, so as to generate high-quality annotation data on a large scale, thereby greatly reducing the cost of manual annotation.

[0033] In order to better understand the above technical solution, exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0034] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0035] Figure 1 FIG. 1 is a flow chart of a method for marking building outlines according to an embodiment of the present application. Figure 1 As shown, the building outline marking method includes the following steps:

[0036] S101, obtaining a plurality of images to be annotated to obtain an original data set.

[0037] As an embodiment, obtaining a plurality of images to be annotated to obtain an original data set includes: obtaining a plurality of remote sensing building images; and processing, cropping and filtering the plurality of remote sensing building images to obtain an original data set.

[0038] That is to say, by collecting a large number of remote sensing building images (high-resolution images), and pre-processing such as processing, cropping and filtering a large number of remote sensing building images to remove unqualified data, and reducing noise through indicator normalization, the original data set can be obtained.

[0039] As a specific embodiment, remote sensing building image data can be collected from relevant urban management agencies, and radiometric and geometric corrections can be performed on high-resolution images. The images are then cropped into a series of 1024*1024 sub-images to fit the input size of the large model (SAM).

[0040] It should be noted that data preprocessing is a key step in data mining, and only high-quality data can form an effective data set.

[0041] S102, manually labeling a portion of the original data set to obtain real labeled data.

[0042] It should be noted that the preprocessed data can be divided into training set, validation set and test set in a ratio of 8:1:1, and 20% of the data can be randomly selected from the training set for manual labeling to generate the real Ground Truth; in order to reduce the cost of manual labeling, a large model can be used to assist in outlining during the labeling process.

[0043] S103, using a first image enhancement method to process a portion of the data set to obtain first enhanced data, and using a second image enhancement method to process a portion of the data set to obtain second enhanced data, and using the first enhanced data, the second enhanced data and a portion of the data set as enhanced data sets.

[0044] That is to say, two different image enhancement methods are used to process a portion of the original data to obtain an enhanced data set.

[0045] As an embodiment, the first image enhancement method is random noise enhancement, and the second image enhancement method is random contrast enhancement.

[0046] Specifically, 20% of the training data were subjected to two different random data enhancements, one for random noise enhancement and the other for random contrast enhancement, and together with the original data formed an enhanced data set of 3 pictures (3*3*1024*1024).

[0047] S104, inputting the enhanced data set into the pre-trained large model to obtain corresponding first predicted annotation data.

[0048] That is to say, the enhanced data is input into the pre-trained large model together with the original data to generate corresponding model annotations; after each sample passes through the large model, it will get three different model annotations, which greatly enriches the diversity of the annotations.

[0049] Specifically, each data set is input into the large model to obtain the corresponding 3 image annotation data (3*1*1024*1024). This data enhancement method can effectively improve the diversity of data.

[0050] S105: Use the first predicted labeled data and the real labeled data to train the conversion network to obtain a trained conversion network.

[0051] That is, the model annotation and manually annotated data are used to train the conversion network. Specifically, the generative adversarial network (GAN) is used for optimization; Figure 2 As shown in the figure, the generator adopts the Unet structure, which is responsible for generating prediction results close to manual annotations based on the model annotations; the discriminator adopts the Resnet structure, which is responsible for distinguishing the generated annotations from the real manual annotations; through generative adversarial training, the generator eventually becomes a conversion network, and the discriminator is used for subsequent data screening to ensure the quality of the annotations.

[0052] As an embodiment, the conversion network is trained using the first predicted labeled data and the real labeled data to obtain a trained conversion network, including: inputting the first predicted labeled data into the generator to obtain the second predicted labeled data; obtaining a first loss function according to the real labeled data and the second predicted labeled data, and obtaining a total loss function of the conversion network according to the first loss function and the second loss function corresponding to the discriminator; and training the generator using the total loss function to obtain a trained conversion network.

[0053] As an embodiment, the total loss function of the conversion network is obtained according to the following formula:

[0054] L=L1+λL GAN

[0055]

[0056] Among them, L represents the total loss function, λ represents the weight parameter of adaptive learning, represents the gradient of the loss function at the last layer of the generator, δ = 10 -6 represents the adjustment factor of the learning rate, x represents the real labeled data, represents the second predicted labeled data, and D represents the discriminator.

[0057] It should be noted that Unet is selected as the conversion network because it can realize the annotation conversion without changing the image size, and its structural characteristics are suitable for processing the building contour segmentation task; the large model annotation data (3*1*1024*1024) is used as the input of Unet, and the network will enhance its robustness through diversified input. At the same time, since these annotation data come from the same original data and have a high degree of semantic association, the network can learn the structural information of the image more easily. The three images of each input data correspond to the three input channels of Unet, and the final output is a single-channel predicted annotation (1*1024*1024).

[0058] Since the large model annotated data already has a certain accuracy, the task of the conversion network is to make more detailed corrections, especially for the optimization of edge details. Therefore, the loss function adopts L1 loss, and the accuracy is improved through pixel-level error comparison. In order to avoid the averaging problem caused by L1 loss (causing blurred prediction results), this application further introduces a generative adversarial network (GAN) for optimization. The conversion network is used as a generator, and the discriminator adopts the Resnet structure (which can be replaced according to actual conditions). The discriminator is used to distinguish between the predicted annotations of the conversion network and the real manual annotations.

[0059] S106, using the pre-trained large model and the trained conversion network to label another part of the original data set to obtain a labeling result.

[0060] That is to say, the remaining 80% of unlabeled data is passed through the pre-trained large model and the trained conversion network to automatically generate high-quality labeling results.

[0061] As an example, after the trained conversion network is fixed in weight, the remaining 80% of the training data is input into the large model, and no data augmentation operation is used in the process. Each input sample will generate 1*1024*1024 large model prediction annotation data, and then the annotation data will be copied in 3 channels to obtain an image of size 3*1024*1024, which will be input into the conversion network with fixed weights to generate the final prediction annotation result.

[0062] In addition, the building outline labeling method also includes: inputting the labeling result into the discriminator to score each labeling data in the labeling result through the discriminator; and filtering out the labeling data with a score lower than a preset threshold to obtain the final labeling data.

[0063] In other words, the generated prediction annotation results are rendered with the original image and manually screened after visualization to ensure their accuracy.

[0064] As an example, the generated prediction labeling results are sent back to the trained discriminator, and the discriminator scores the authenticity of each labeled data. Data with a score lower than a set threshold (such as 0.3) given by the discriminator will be discarded, that is, the labeled data with a low similarity to the real label will be excluded to ensure the quality of the labeled data.

[0065] Compared with traditional labeling methods, the building outline labeling method of the disclosed embodiment greatly reduces the labeling cost and time, especially in the field of high-resolution remote sensing images, and can demonstrate excellent performance in large-scale data labeling; thus, it can not only provide accurate data support for urban planning and land and resources management, but can also be widely used in disaster monitoring, environmental assessment and other scenarios, providing strong technical support for various construction and management projects.

[0066] In order to implement the above embodiment, this embodiment also provides a building outline segmentation method.

[0067] Figure 3 is a flow chart of a building outline segmentation method according to an exemplary embodiment. Figure 3 As shown, the method comprises the following steps:

[0068] S201, using the above-mentioned annotation method to obtain the annotation result and its corresponding original image.

[0069] S202, inputting the annotation result and its corresponding original image into the segmentation model to be trained for training, so as to obtain a trained segmentation model.

[0070] S203, obtaining a target image to be segmented.

[0071] S204: Input the target image to be segmented into the trained segmentation module to obtain a corresponding segmentation result.

[0072] That is to say, the final annotation data of the large model annotation and its corresponding original image obtained by the above-mentioned annotation method, as well as the manually annotated real annotation data and its corresponding original image are input into the downstream segmentation model (such as PP-LiteSeg or SegFormer) to complete the final segmentation network training process, thereby achieving accurate building contour segmentation tasks.

[0073] The building outline segmentation method of the disclosed embodiment obtains the labeling result by utilizing the above-mentioned building outline labeling method, thereby greatly reducing the cost of manual labeling and improving the performance of the segmentation model.

[0074] In order to implement the above embodiment, the embodiment of the present disclosure provides a building outline marking device.

[0075] Figure 4 FIG. 1 is a block diagram of a device for marking building outlines according to an exemplary embodiment. Figure 4 The device includes: a first acquisition module 1, a manual labeling module 2, a data processing module 3, a prediction labeling module 4, a conversion network training module 5 and an optimization labeling module 6.

[0076] Among them, the first acquisition module 1 is used to acquire multiple images to be annotated to obtain the original data set; the manual annotation module 2 is used to manually annotate a part of the original data set to obtain real annotated data; the data processing module 3 is used to process a part of the data set using the first image enhancement method to obtain first enhanced data, and to process a part of the data set using the second image enhancement method to obtain second enhanced data, and use the first enhanced data, the second enhanced data and a part of the data set as the enhanced data set; the prediction and annotation module 4 is used to input the enhanced data set into the pre-trained large model to obtain the corresponding first predicted annotation data; the conversion network training module 5 is used to train the conversion network using the first predicted annotation data and the real annotation data to obtain a trained conversion network; the optimization annotation module 6 is used to annotate another part of the data set in the original data set using the pre-trained large model and the trained conversion network to obtain the annotation result.

[0077] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the building outline marking method, and will not be elaborated here.

[0078] In summary, according to the building outline labeling device of the embodiment of the present application, by combining manual labeling and large model labeling, the mapping between the two is realized with the help of a conversion network, so as to generate high-quality labeling data on a large scale, thereby greatly reducing the cost of manual labeling.

[0079] In order to implement the above embodiment, the embodiment of the present disclosure provides a building outline segmentation device.

[0080] Figure 5 is a block diagram of a building outline segmentation device according to an exemplary embodiment. Figure 5 The device includes: a second acquisition module 10, a segmentation model training module 20, a third acquisition module 30 and a segmentation module 40.

[0081] Among them, the second acquisition module 10 is used to adopt the above-mentioned annotation method to obtain the annotation results and the corresponding original images; the segmentation model training module 20 is used to input the annotation results and the corresponding original images into the segmentation model to be trained for training to obtain a trained segmentation model; the third acquisition module 40 is used to acquire the target image to be segmented; the segmentation module 50 is used to input the target image to be segmented into the trained segmentation module to obtain the corresponding segmentation result.

[0082] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the building outline segmentation method, and will not be elaborated here.

[0083] In summary, according to the building outline segmentation device of the embodiment of the present application, the labeling result is obtained by utilizing the above-mentioned building outline labeling method, thereby greatly reducing the cost of manual labeling and improving the performance of the segmentation model.

[0084] In order to implement the above embodiment, the embodiment of the present disclosure provides a computer-readable storage medium, on which a building outline labeling program is stored. When the building outline labeling program is executed, it implements the above-mentioned building outline labeling method.

[0085] According to the computer-readable storage medium of an embodiment of the present invention, a building outline labeling program is used so that the processor implements the building outline labeling method as described above when executing the building outline labeling program. Thus, by combining manual labeling and large model labeling, the mapping between the two is achieved with the help of a conversion network, so as to generate high-quality labeling data on a large scale, thereby greatly reducing the cost of manual labeling.

[0086] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0087] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0088] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0089] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0090] It should be noted that in the claims, any reference signs placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The present application may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names.

[0091] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0092] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

[0093] In the description of the present application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the feature. In the description of the present application, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0094] In this application, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0095] In the present application, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0096] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms should not be understood as necessarily being directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0097] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for marking building outlines, characterized in that: The following steps are involved: Obtain multiple images to be labeled to obtain an original data set; Manually labeling a portion of the original data set to obtain real labeled data; Using a first image enhancement method to process the portion of the data set to obtain first enhanced data, and using a second image enhancement method to process the portion of the data set to obtain second enhanced data, and using the first enhanced data, the second enhanced data, and the portion of the data set as enhanced data sets; Inputting the enhanced data set into the pre-trained large model to obtain corresponding first predicted labeled data; Using the first predicted labeled data and the real labeled data to train a conversion network to obtain a trained conversion network; The pre-trained large model and the trained conversion network are used to label another part of the original data set to obtain a labeling result.

2. The building outline marking method according to claim 1, characterized in that: Get multiple images to be annotated to get the original data set, including: Acquire multiple remote sensing building images; The multiple remote sensing building images are processed, cropped and filtered to obtain an original data set.

3. The building outline marking method according to claim 1, characterized in that: The first image enhancement method is random noise enhancement, and the second image enhancement method is random contrast enhancement.

4. The building outline marking method according to claim 1, characterized in that: The conversion network is trained using the first predicted labeled data and the real labeled data to obtain a trained conversion network, including: Inputting the first predicted labeled data into a generator to obtain second predicted labeled data; Obtaining a first loss function according to the real labeled data and the second predicted labeled data, and obtaining a total loss function of the conversion network according to the first loss function and a second loss function corresponding to the discriminator; The generator is trained using the total loss function to obtain a trained conversion network.

5. The building outline marking method according to claim 4, characterized in that: The total loss function of the conversion network is obtained according to the following formula: L=L1+λL GAN Among them, L represents the total loss function, λ represents the weight parameter of adaptive learning, represents the gradient of the loss function at the last layer of the generator, δ = 10 -6 represents the adjustment factor of the learning rate, x represents the real labeled data, represents the second predicted labeled data, and D represents the discriminator.

6. The building outline marking method according to claim 5, characterized in that: Also includes: Inputting the labeling result into the discriminator so as to score each labeling data in the labeling result through the discriminator; The labeled data with scores lower than the preset threshold are filtered out to obtain the final labeled data.

7. A building outline segmentation method, characterized in that: The following steps are involved: Using the labeling method according to any one of claims 1 to 6 to obtain a labeling result and its corresponding original image; Inputting the annotation results and the corresponding original images into the segmentation model to be trained to obtain a trained segmentation model; Obtain a target image to be segmented; The target image to be segmented is input into the trained segmentation module to obtain a corresponding segmentation result.

8. A building outline marking device, characterized in that: include: A first acquisition module is used to acquire a plurality of images to be annotated to obtain an original data set; A manual labeling module, used to manually label a portion of the original data set to obtain real labeled data; a data processing module, configured to process the part of the data set using a first image enhancement method to obtain first enhanced data, and process the part of the data set using a second image enhancement method to obtain second enhanced data, and use the first enhanced data, the second enhanced data and the part of the data set as enhanced data sets; A prediction and annotation module, used for inputting the enhanced data set into the pre-trained large model to obtain corresponding first prediction and annotation data; A conversion network training module, used to train a conversion network using the first predicted labeled data and the real labeled data to obtain a trained conversion network; The optimized labeling module is used to label another part of the original data set by using the pre-trained large model and the trained conversion network to obtain a labeling result.

9. A building outline segmentation device, characterized in that: include: A second acquisition module, configured to obtain a labeling result and a corresponding original image by using the labeling method according to any one of claims 1 to 6; A segmentation model training module, used for inputting the annotation results and the corresponding original images into the segmentation model to be trained to obtain a trained segmentation model; A third acquisition module is used to acquire a target image to be segmented; The segmentation module is used to input the target image to be segmented into the trained segmentation module to obtain a corresponding segmentation result.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a building outline marking program, and the building outline marking program implements the building outline marking method according to any one of claims 1 to 6 when executed.

Images