A deep learning-based multi-task building extraction and change detection method

Abstract

The application provides a kind of multi-task building extraction and change detection method based on deep learning, joint building extraction and change detection task through deep neural network architecture, including: encoder network, building extraction decoder network and building change decoder network, to realize the identification task of building in remote sensing image, and detect building change area task;Multi-task building extraction and change detection neural network is used to preprocess satellite remote sensing image data based on multi-phase and generate different time phase image pairs in the data set as the data input of two channels of image T1 and image T2 respectively, and the deep features of building in the image are extracted through the encoder network respectively, the features extracted by image T1 channel are obtained through building extraction decoder network to obtain building extraction result, and the features extracted by image T1 channel and the features extracted by image T2 channel are obtained after building change decoder network to obtain building change detection result.

Description

Technical Field

[0001] This invention relates to the fields of deep learning, computer vision, building extraction, and change detection, and in particular to a multi-task building extraction and change detection method based on deep learning. Background Technology

[0002] In the power grid sector, building extraction and change detection technologies based on remote sensing images have wide applications, particularly in power facility monitoring, line inspection, and post-disaster assessment. Traditional building extraction and change detection methods typically rely on manually designed feature extractors and classifiers, which suffer from inefficiency and low accuracy when processing large-scale remote sensing data. Deep learning algorithms, with their powerful feature learning capabilities and adaptability, have become an effective tool for solving building extraction and change detection problems. Deep learning-based methods can automatically learn feature representations from images without requiring manual feature design, and the model can be optimized through end-to-end training. These advantages have led to significant performance improvements in building extraction and change detection tasks using deep learning-based methods.

[0003] However, existing deep learning-based methods for building extraction and change detection still have some problems. Traditional methods typically treat building extraction and change detection as two separate tasks, ignoring the correlation between them. Summary of the Invention

[0004] This invention employs deep learning to construct a deep neural network architecture that combines building extraction and change detection tasks, achieving efficient multi-task processing to address the limitation of existing technologies that cannot simultaneously perform both tasks. The invention designs a multi-task building extraction and change detection neural network, including an encoder network, a building extraction decoder network, and a building change decoder network. This network can accurately identify buildings in remote sensing imagery and effectively detect areas of building change. This multi-task learning method not only improves detection accuracy and efficiency but also fully utilizes complementary information between tasks, enhancing the model's generalization ability. It is suitable for practical applications such as power transmission line corridor environmental monitoring, urban planning, disaster assessment, and land use monitoring.

[0005] The present invention specifically adopts the following technical solution:

[0006] A deep learning-based multi-task building extraction and change detection method: This method combines building extraction and change detection tasks using a deep neural network architecture, including an encoder network, a building extraction decoder network, and a building change decoder network, to achieve the task of identifying buildings in remote sensing images and detecting areas of building change.

[0007] A multi-task building extraction and change detection neural network is used to preprocess satellite remote sensing image data based on multiple time phases and generate image pairs of different time phases in the dataset as data inputs for image T1 and image T2 channels respectively. The deep features of buildings in the images are extracted by encoder network. The features extracted from image T1 channel are passed to building extraction decoder network to obtain building extraction results. The features extracted from image T1 channel and image T2 channel are passed to building change decoder network to obtain building change detection results.

[0008] Furthermore, data preprocessing includes: collecting satellite remote sensing image data from multiple time periods, performing radiometric, atmospheric, and geometric corrections on the images; then cropping the images according to the target area range to remove irrelevant areas; and finally registering the multiple time period images to ensure that the same location is aligned in images from different time periods by having the same spatial reference frame.

[0009] The building extraction and change detection dataset is established by annotating buildings in preprocessed images to generate building outline data; comparing images from different time periods, annotating the areas of increase, decrease and change of buildings, and generating annotation data for change detection.

[0010] Furthermore, the encoder network includes a CNN feature extraction module and a multi-scale feature extraction module;

[0011] The CNN feature extraction module consists of a 7×7 convolutional layer with 64 filters and a stride of 2, a 3×3 max pooling layer with a stride of 2, 3 convolutional blocks 1, 4 convolutional blocks 2, and 6 convolutional blocks 3. Convolutional block 1 contains 3 residual blocks, each containing 3 convolutional layers: a 1×1 convolution with 64 filters, a 3×3 convolution with 64 filters, and a 1×1 convolution with 256 filters. Convolutional block 2... Convolutional block 2 contains four residual blocks, each containing three convolutional layers: a 1×1 convolution with 256 filters, a 3×3 convolution with 256 filters, and a 1×1 convolution with 1024 filters. Convolutional block 3 contains six residual blocks, each containing three convolutional layers: a 1×1 convolution with 256 filters, a 3×3 convolution with 256 filters, and a 1×1 convolution with 1024 filters. Convolutional blocks 2 and 3 output low-dimensional features L. low and high-dimensional features L high After passing through the CNN feature extraction module, the T1 and T2 image channels respectively obtain feature L. low1 L high1 and feature L low2 L high2 ;

[0012] The multi-scale feature extraction module uses a 3×3 convolutional structure to process objects of different scales in building extraction and change detection tasks. The module first performs dimensionality reduction through 1×1 convolution and then 3×3 convolution to obtain a 3×3 receptive field. The right branch then performs 1×1 convolution and 3×3 convolution to obtain a 5×5 receptive field. The left branch then performs 1×1 convolution and 3×3 convolution to obtain a 7×7 receptive field. The three receptive fields of different sizes are fused to capture feature information of different scales.

[0013] Furthermore, the building extraction decoder network extracts the low-dimensional features L from the T1 channel of the image via the CNN feature extraction module. low1 Next, we will focus on the low-latitude feature L. low1 The results obtained through 1×1 convolution, batch normalization, and ReLU activation function layers are compared with the high-dimensional features L output by the CNN feature extraction module. high1 After the multi-scale feature extraction module and the results of 4x upsampling are superimposed, feature L is obtained. code1 , feature L code1 After further processing with 3×3 convolution, batch normalization, ReLU activation function layers, and 4x upsampling, the building extraction results are obtained.

[0014] Furthermore, the building change decoder network will extract the low-dimensional features L from the CNN feature extraction module of the T2 channel of the image. low2 The result obtained after 1×1 convolution, batch normalization, and ReLU activation function layers is compared with the high-dimensional feature L output by the CNN feature extraction module. high2 After the results of the multi-scale feature extraction module and 4x upsampling are superimposed, feature L is obtained. code2 Then, feature L code2 The feature L obtained above code1 The connections are made, and the results are then processed through a 3×3 convolution, batch normalization, and ReLU activation function layer to obtain the fused feature L. F1 ; feature L high1 and L high2 The features extracted by the multi-scale feature extraction module are concatenated, and the result is combined with the fused feature L obtained after 3×3 convolution, batch normalization, ReLU activation function layer, and 4x upsampling. F2 ; feature L F1 and feature L F2 After concatenation, the data is processed through a 3×3 convolution, batch normalization, ReLU activation function layer, and 4x upsampling to obtain the building change detection results.

[0015] Furthermore, the loss function of the multi-task building extraction and change detection neural network includes a building extraction loss function and a change detection loss function:

[0016] The loss for building extraction is calculated using the binary cross-entropy loss function:

[0017]

[0018] in, It is the true label of the i-th pixel. is the probability that the i-th pixel is predicted to be a building, and N is the total number of pixels;

[0019] The loss for change detection is calculated using the binary cross-entropy loss function:

[0020]

[0021] in, It is the true label of the change in the i-th pixel. is the probability that the i-th pixel is predicted to be a building change, and N is the total number of pixels;

[0022] The loss functions of the two tasks are summed with weights. and Used to balance the losses of the two tasks:

[0023] .

[0024] Furthermore, the process of training the multi-task building extraction and change detection neural network is as follows: The building extraction and change detection dataset is used as the input to the multi-task building extraction and change detection neural network, and the outline data of the labeled buildings and the labeled change areas of the buildings are used as the output to train the multi-task building extraction and change detection neural network; PyTorch is used as the deep neural network learning framework for model training, the Adam optimizer is used, the initial learning rate is set to 0.01, and a hierarchical descent learning rate method is adopted, with 200 epochs of training; two registered satellite images from different time phases are input into the trained neural network to obtain the building regions and their changes in the satellite images.

[0025] A deep learning-based multi-task building extraction and change detection system includes:

[0026] The preprocessing module is used to preprocess multi-temporal satellite remote sensing image data and generate datasets;

[0027] In addition, a multi-task building extraction and change detection neural network, including an encoder network, a building extraction decoder network, and a building change decoder network, is used to realize the task of identifying buildings in remote sensing images and the task of detecting changes in building areas;

[0028] A multi-task building extraction and change detection neural network is used to preprocess satellite remote sensing image data based on multiple time phases and generate image pairs of different time phases in the dataset as data inputs for image T1 and image T2 channels respectively. The deep features of buildings in the images are extracted by encoder network. The features extracted from image T1 channel are passed to building extraction decoder network to obtain building extraction results. The features extracted from image T1 channel and image T2 channel are passed to building change decoder network to obtain building change detection results.

[0029] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a deep learning-based multi-task building extraction and change detection method as described above.

[0030] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a deep learning-based multi-task building extraction and change detection method as described above.

[0031] Compared to existing technologies, this invention and its preferred embodiment employ deep learning methods. By constructing a deep neural network architecture, it combines building extraction and change detection tasks to achieve efficient multi-task processing, thus solving the problem that existing technologies cannot simultaneously perform both building extraction and change detection tasks. This invention designs a multi-task building extraction and change detection neural network, including an encoder network, a building extraction decoder network, and a building change decoder network. This network can accurately identify buildings in remote sensing imagery and effectively detect areas of building change. This multi-task learning method not only improves detection accuracy and efficiency but also fully utilizes complementary information between tasks, enhancing the model's generalization ability. It is suitable for practical applications such as power transmission line corridor environmental monitoring, urban planning, disaster assessment, and land use monitoring. Attached Figure Description

[0032] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0033] Figure 1 This is a diagram of the neural network structure for multi-task building extraction and change detection according to an embodiment of the present invention.

[0034] Figure 2 This is a structural diagram of the CNN feature extraction module according to an embodiment of the present invention;

[0035] Figure 3 This is a structural diagram of the multi-scale feature extraction module according to an embodiment of the present invention;

[0036] Figure 4This is a schematic diagram illustrating the implementation process of an embodiment of the present invention. Detailed Implementation

[0037] In the following, specific embodiments of this application will be described in detail with reference to the accompanying drawings. Based on these detailed descriptions, those skilled in the art will be able to clearly understand and implement this application. Without departing from the principles of this application, features from various embodiments can be combined to obtain new implementations, or certain features from some embodiments can be substituted to obtain other preferred implementations.

[0038] To make the features and advantages of this patent more apparent and understandable, specific embodiments are provided below, along with accompanying drawings, for detailed explanation:

[0039] like Figure 4 As shown in the figure, this invention proposes a multi-task building extraction and change detection method based on deep learning, which can simultaneously complete the two tasks of building extraction and change detection. By designing a deep neural network architecture, the building extraction and change detection tasks are combined, making full use of the correlation between the two to improve the accuracy and robustness of the model.

[0040] The implementation process can refer to the following steps:

[0041] Step S1: Data Preprocessing. Collect multi-temporal satellite remote sensing image data and perform radiometric, atmospheric, and geometric corrections on the images to eliminate sensor noise, atmospheric effects, and geometric distortion, giving the images uniform spectral and geometric characteristics. Then, based on the target area, crop the images to remove irrelevant regions, reduce data volume, and register the multi-temporal images to ensure they have the same spatial reference frame, guaranteeing alignment of the same location in images from different time periods, facilitating change detection.

[0042] Step S2: Establish a building extraction and change detection dataset. Manually label the buildings in the images collected and processed in Step S1 to generate building outline data; compare images from different time periods to label areas of increase, decrease, and change in buildings, generating change detection annotation data; check the matching degree between the annotation data and the images to ensure the accuracy and consistency of the annotations.

[0043] Step S3: Construct a multi-task building extraction and change detection neural network. For example... Figure 1As shown, the multi-task building extraction and change detection neural network includes an encoder network, a building extraction decoder network, and a building change decoder network. The image pairs obtained in step S2 at different times are used as data inputs for image T1 and image T2, respectively. The encoder network extracts deep-level features of buildings in the images. The features extracted from image T1 are then processed by the building extraction decoder network to obtain the building extraction result. Finally, the features extracted from image T1 and image T2 are processed by the building change decoder network to obtain the building change detection result.

[0044] The decoder network from step S3 is then constructed. The decoder network includes a CNN feature extraction module and a multi-scale feature extraction module.

[0045] CNN feature extraction module, such as Figure 2 The diagram shows a convolutional layer consisting of a 7×7 convolution with 64 filters and a stride of 2, a 3×3 max pooling layer with a stride of 2, and three convolutional blocks (1), four convolutional blocks (2), and six convolutional blocks (3). Convolutional block 1 contains three residual blocks, each containing three convolutional layers: a 1×1 convolution (64 filters), a 3×3 convolution (64 filters), and a 1×1 convolution (256 filters). Convolutional block 2 contains four residual blocks, each containing three convolutional layers: a 1×1 convolution (256 filters), a 3×3 convolution (256 filters), and a 1×1 convolution (1024 filters). Convolutional block 3 contains six residual blocks, each containing three convolutional layers: a 1×1 convolution (256 filters), a 3×3 convolution (256 filters), and a 1×1 convolution (1024 filters). Convolutional block 2 and convolutional block 3 output low-dimensional features L respectively. low and high-dimensional features L high After passing through the CNN feature extraction module, the T1 and T2 image channels respectively obtain feature L. low1 L high2 and feature L low2 L high2 .

[0046] The core of the multi-scale feature extraction module is a 3×3 convolutional structure, such as... Figure 3 As shown, this is beneficial for handling objects of different scales in building extraction and change detection tasks. This module can effectively aggregate receptive fields of 3×3, 5×5, and 7×7 sizes. First, a 3×3 receptive field is obtained through 1×1 convolution for dimensionality reduction and then a 3×3 convolution; the right branch then undergoes another 1×1 and 3×3 convolution to obtain a 5×5 receptive field; the left branch then undergoes another 1×1 and 3×3 convolution to obtain a 7×7 receptive field. Fusing these three different receptive field sizes allows for the capture of feature information at different scales.

[0047] The building extraction decoder network described in further construction step S3. (As...) Figure 1 As shown, the low-dimensional feature L output by the CNN feature extraction module is obtained by processing the T1 channel of the image. low1 Next, we will focus on the low-latitude feature L. low1 The results obtained through 1×1 convolution, batch normalization, and ReLU activation layers are compared with the high-dimensional features L output by the CNN feature extraction module. high1 After the multi-scale feature extraction module and the results of 4x upsampling are superimposed, feature L is obtained. code1 , feature L code1 After further processing with 3×3 convolution, batch normalization, ReLU activation function layers, and 4x upsampling, the building extraction results are obtained.

[0048] The building change decoder network described in step S3 is further constructed. (As...) Figure 1 As shown, the low-dimensional feature L output by the CNN feature extraction module of the T2 channel of the image is... low2 The results obtained after 1×1 convolution, batch normalization, and ReLU activation function layers are compared with the high-dimensional features L output by the CNN feature extraction module. high2 After the results of the multi-scale feature extraction module and 4x upsampling are superimposed, feature L is obtained. code2 Then, feature L code2 The feature L obtained above code1 The connection is performed, and the result is passed through a 3×3 convolution, batch normalization, and ReLU activation function layer to obtain the fused feature L. F1 ; feature L high1 and L high2 The features extracted by the multi-scale feature extraction module are concatenated, and the result is combined with the fused feature L obtained after 3×3 convolution, batch normalization, ReLU activation function layer, and 4x upsampling. F2 ; feature L F1 and feature L F2 After concatenation, the data is processed through a 3×3 convolution, batch normalization, and ReLU activation function layer, and then upsampled by 4 times to obtain the results of building change detection.

[0049] Step S4: Construct the loss function for the multi-task building extraction and change detection neural network. The loss function for the multi-task building extraction and change detection neural network includes a building extraction loss function and a change detection loss function.

[0050] Building extraction is a binary classification problem (buildings and non-buildings). The loss for building extraction can be calculated using the binary cross-entropy loss function:

[0051]

[0052] in, It is the true label of the i-th pixel. is the probability that the i-th pixel is predicted to be a building, and N is the total number of pixels.

[0053] Change detection loss is a binary classification problem (change and no change). A similar binary cross-entropy loss function can be used to calculate the loss for change detection:

[0054]

[0055] in, It is the true label of the change in the i-th pixel. is the probability that the i-th pixel is predicted to be a building change, and N is the total number of pixels.

[0056] To construct a joint loss function for multiple tasks, the loss functions of two tasks can be summed with weights. and Used to balance the losses of the two tasks:

[0057]

[0058] Step S5: Train the multi-task building extraction and change detection neural network. Use the building extraction and change detection dataset established in Step S2 as the input to the multi-task building extraction and change detection neural network, and use manually labeled building outline data and labeled building change regions as the output to train the multi-task building extraction and change detection neural network. Use PyTorch as the deep neural network learning framework and NVIDIA GeForce RTX4090 as the GPU for model training. Use the Adam optimizer, set the initial learning rate to 0.01, and employ a hierarchical descent learning rate method. The training iterations are 200 epochs (all training data completes one computation).

[0059] Step S6: Identify building areas and their changes in satellite imagery. Input the two registered satellite images from different time periods into the neural network trained in step S5 to obtain the results of building areas and their changes in the satellite imagery.

[0060] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

[0066] This patent is not limited to the above-described preferred embodiment. Anyone can derive other forms of deep learning-based multi-task building extraction and change detection methods under the guidance of this patent. All equivalent changes and modifications made within the scope of this patent application shall fall within the scope of this patent.

Claims

1. A multi-task building extraction and change detection method based on deep learning, characterized in that: By combining building extraction and change detection tasks through a deep neural network architecture, including an encoder network, a building extraction decoder network, and a building change decoder network, the task of identifying buildings in remote sensing images and detecting areas of building change can be achieved. A multi-task building extraction and change detection neural network is used to preprocess satellite remote sensing image data based on multiple time phases and generate image pairs of different time phases in the dataset as data inputs for image T1 and image T2 channels respectively. The deep features of buildings in the images are extracted by encoder network. The features extracted from image T1 channel are passed to building extraction decoder network to obtain building extraction results. The features extracted from image T1 channel and image T2 channel are passed to building change decoder network to obtain building change detection results. The encoder network includes a CNN feature extraction module and a multi-scale feature extraction module; The CNN feature extraction module consists of a 7×7 convolutional layer with 64 filters and a stride of 2, a 3×3 max pooling layer with a stride of 2, 3 convolutional blocks 1, 4 convolutional blocks 2, and 6 convolutional blocks 3. Convolutional block 1 contains 3 residual blocks, each containing 3 convolutional layers: a 1×1 convolution with 64 filters, a 3×3 convolution with 64 filters, and a 1×1 convolution with 256 filters. Convolutional block 2... Convolutional block 2 contains four residual blocks, each containing three convolutional layers: a 1×1 convolution with 256 filters, a 3×3 convolution with 256 filters, and a 1×1 convolution with 1024 filters. Convolutional block 3 contains six residual blocks, each containing three convolutional layers: a 1×1 convolution with 256 filters, a 3×3 convolution with 256 filters, and a 1×1 convolution with 1024 filters. Convolutional blocks 2 and 3 output low-dimensional features L. low and high-dimensional features L high After the T1 channel of the image is processed by the CNN feature extraction module, feature L is obtained respectively. low1 L high1 After the T2 channel of the image is processed by the CNN feature extraction module, feature L is obtained respectively. low2 L high2 ; The multi-scale feature extraction module uses a 3×3 convolutional structure to process objects of different scales in building extraction and change detection tasks. The module first performs dimensionality reduction through 1×1 convolution and then 3×3 convolution to obtain a 3×3 receptive field. The right branch then performs 1×1 convolution and 3×3 convolution to obtain a 5×5 receptive field. The left branch then performs 1×1 convolution and 3×3 convolution to obtain a 7×7 receptive field. The three receptive fields of different sizes are fused to capture feature information of different scales. The building extraction decoder network extracts the low-dimensional features L from the T1 channel of the image via the CNN feature extraction module. low1 Next, we will focus on the low-latitude feature L. low1 The results obtained through 1×1 convolution, batch normalization, and ReLU activation function layers are compared with the high-dimensional features L output by the CNN feature extraction module. high1 After the multi-scale feature extraction module and the results of 4x upsampling are superimposed, feature L is obtained. code1 , feature L code1 After further processing with 3×3 convolution, batch normalization, ReLU activation function layers, and 4x upsampling, the building extraction results are obtained. The building change decoder network will extract the low-dimensional features L from the CNN feature extraction module of the T2 channel of the image. low2 The result obtained after 1×1 convolution, batch normalization, and ReLU activation function layers is compared with the high-dimensional feature L output by the CNN feature extraction module. high2 After the results of the multi-scale feature extraction module and 4x upsampling are superimposed, feature L is obtained. code2 Then, feature L code2 The feature L obtained above code1 The connections are made, and the results are then processed through a 3×3 convolution, batch normalization, and ReLU activation function layer to obtain the fused feature L. F1 ; feature L high1 and L high2 The features extracted by the multi-scale feature extraction module are concatenated, and the result is combined with the fused feature L obtained after 3×3 convolution, batch normalization, ReLU activation function layer, and 4x upsampling. F2 ; feature L F1 and feature L F2 After concatenation, the data is processed through a 3×3 convolution, batch normalization, ReLU activation function layer, and 4x upsampling to obtain the building change detection results.

2. The method for multi-task building extraction and change detection based on deep learning according to claim 1, characterized in that: Data preprocessing includes: collecting satellite remote sensing image data from multiple time periods, performing radiometric, atmospheric, and geometric corrections on the images; then cropping the images according to the target area range to remove irrelevant areas; and finally registering the multiple time period images to ensure that the same location is aligned in images from different time periods by having the same spatial reference frame. The building extraction and change detection dataset is established by annotating buildings in preprocessed images to generate building outline data; comparing images from different time periods, annotating the areas of increase, decrease and change of buildings, and generating annotation data for change detection.

3. The method for multi-task building extraction and change detection based on deep learning according to claim 1, characterized in that: The loss function of the multi-task building extraction and change detection neural network includes a building extraction loss function and a change detection loss function: The loss for building extraction is calculated using the binary cross-entropy loss function: in, It is the true label of the i-th pixel. is the probability that the i-th pixel is predicted to be a building, and N is the total number of pixels; The loss for change detection is calculated using the binary cross-entropy loss function: in, It is the true label of the change in the i-th pixel. is the probability that the i-th pixel is predicted to be a building change, and N is the total number of pixels; The loss functions of the two tasks are summed with weights. and Used to balance the losses of the two tasks: 。 4. The method for multi-task building extraction and change detection based on deep learning according to claim 1, characterized in that: The process of training a multi-task building extraction and change detection neural network is as follows: the building extraction and change detection dataset is used as the input of the multi-task building extraction and change detection neural network, and the outline data of the labeled buildings and the labeled building change areas are used as the output to train the multi-task building extraction and change detection neural network. PyTorch was used as the deep neural network learning framework to train the model. The Adam optimizer was used with an initial learning rate of 0.01 and a hierarchical descent learning rate method was adopted. The training was performed for 200 epochs. Two registered satellite images from different time periods were input into the trained neural network to obtain the building areas and their changes in the satellite images.

5. A deep learning-based multi-task building extraction and change detection system, used to implement the method as described in claim 1, characterized in that, include: The preprocessing module is used to preprocess multi-temporal satellite remote sensing image data and generate datasets; In addition, a multi-task building extraction and change detection neural network, including an encoder network, a building extraction decoder network, and a building change decoder network, is used to realize the task of identifying buildings in remote sensing images and the task of detecting changes in building areas; A multi-task building extraction and change detection neural network is used to preprocess satellite remote sensing image data based on multiple time phases and generate image pairs of different time phases in the dataset as data inputs for image T1 and image T2 channels respectively. The deep features of buildings in the images are extracted by encoder network. The features extracted from image T1 channel are passed to building extraction decoder network to obtain building extraction results. The features extracted from image T1 channel and image T2 channel are passed to building change decoder network to obtain building change detection results.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the deep learning-based multi-task building extraction and change detection method as described in any one of claims 1-4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of a deep learning-based multi-task building extraction and change detection method as described in any one of claims 1-4.

Images