Remote sensing building extraction method and device

Abstract

The present invention provides a remote sensing building extraction method and device, wherein the method comprises: obtaining a remote sensing image to be extracted; inputting the remote sensing image to be extracted into an improved D-linknet model to obtain a corresponding extraction result; post-processing the extraction result to obtain a corresponding post-processing result; fusing the post-processing result with the remote sensing image to be extracted to obtain a final extraction result. The present invention can improve the extraction accuracy by using the improved D-linknet model to extract buildings, and can remove isolated points by post-processing the extraction results, so that the extraction results are more accurate.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a remote sensing building extraction method and device. Background Art

[0002] With the rapid development of artificial intelligence technology, especially machine learning technology, traditional building extraction methods often have problems such as complex processes, low degree of automation, and limited promotion capabilities. Some research institutions or scholars have tried to use artificial intelligence technology, especially deep learning technology, to study building target extraction and have achieved certain results. [1] Remote sensing building extraction is actually a binary semantic segmentation problem. The commonly used network in this field is the Unet series. The Unet series is a series of improved algorithms based on Unet. The U-Net model uses an encoder structure to implement layer-by-layer feature extraction and a decoder to gradually restore contextual detail information. U-Net uses deconvolution operations instead of linear interpolation operations, and uses jump connections to fuse feature maps in the encoder network, thereby ensuring that the decoder network can receive more contextual detail features.

[0003] The original U-Net only has four downsampling processes, the overall receptive field of the network is small, and the input and output sizes of the network have great limitations, which makes the network ineffective for many high-resolution images. Therefore, a common improvement to U-Net is to directly stack several convolution modules and pooling layers on the basis of U-Net. Compared with the original U-Net, the network can integrate more semantic information in space. However, although the deepened U-Net can solve the problem of the original U-Net's small receptive field, its detailed information is not well preserved. Summary of the invention

[0004] The present invention provides a remote sensing building extraction method and system, which are used to solve the defect of insufficient building extraction accuracy in the prior art and realize accurate extraction of buildings.

[0005] In a first aspect, the present invention provides a remote sensing building extraction method, comprising:

[0006] Obtaining remote sensing images to be extracted;

[0007] Inputting the remote sensing image to be extracted into the improved D-linknet model to obtain the corresponding extraction result;

[0008] Post-processing the extraction result to obtain a corresponding post-processing result;

[0009] The post-processing result is fused with the remote sensing image to be extracted to obtain the final extraction result.

[0010] Further, according to the remote sensing building extraction method provided by the present invention, the improved D-linknet model includes a new encoding module, a new hole convolution module, a first channel attention module, a second channel attention module and a decoding module:

[0011] The new encoding module is obtained by adding at least one skip-layer connection in the hierarchy of the encoding module of the original D-linknet model;

[0012] The new dilated convolution module is obtained by adjusting the dilated rate of the dilated convolution module of the original D-linknet model;

[0013] The first channel attention module is between the new encoding module and the new hole convolution module, and the second channel attention module is between the new hole convolution module and the decoding module.

[0014] Further, according to the remote sensing building extraction method provided by the present invention, the improved D-linknet model parameters are trained by the following method:

[0015] Step 310: Obtain a remote sensing data set for training;

[0016] Step 320: preprocessing the remote sensing data set for training to obtain a preprocessed remote sensing data set;

[0017] Step 330: Divide the processed remote sensing data set into a training data set and a test data set;

[0018] Step 340: inputting the training data set into the improved D-linknet model for training until the corresponding loss function converges, thereby obtaining an improved D-linknet model with corresponding parameters;

[0019] Step 350: input the test data set into the improved D-linknet model of the corresponding parameters to obtain the corresponding accuracy;

[0020] Step 360: Compare the accuracy with a preset accuracy;

[0021] Step 370: if the accuracy is greater than the preset accuracy, re-execute step 340; if the accuracy is not greater than the preset accuracy, execute step 380;

[0022] Step 380: Stop training and fix the improved D-linknet model parameters.

[0023] Further, according to the remote sensing building extraction method provided by the present invention, the preprocessing of the remote sensing data set used for training to obtain the preprocessed remote sensing data set specifically includes:

[0024] Vectorizing the buildings in the remote sensing image in the remote sensing data set to obtain a vector layer;

[0025] Convert the vectorized layer into a raster layer to generate a label image;

[0026] The remote sensing image and the corresponding label image are cropped to the same size to obtain a processed remote sensing data set.

[0027] Further, according to the remote sensing building extraction method provided by the present invention, the step of inputting the training data set into the improved D-linknet model for training comprises:

[0028] During the training process, data enhancement is performed on each image in the training data set to obtain an enhanced image;

[0029] The improved D-linknet model is trained according to the enhanced image.

[0030] Further, according to the remote sensing building extraction method provided by the present invention, the post-processing of the extraction result to obtain the corresponding post-processing result specifically includes:

[0031] Performing isolated point removal processing on the extraction result to obtain an extraction result after the isolated points are removed;

[0032] The extraction result after removing isolated points is subjected to a minimum circumscribed rectangle optimization operation to obtain a corresponding post-processing result.

[0033] In a second aspect, the present invention provides a remote sensing building extraction device, comprising:

[0034] A remote sensing image acquisition module is used to acquire remote sensing images to be extracted;

[0035] A result extraction module, used for inputting the remote sensing image to be extracted into the improved D-linknet model to obtain the corresponding extraction result;

[0036] A post-processing module, specifically used to post-process the extraction result to obtain a corresponding post-processing result;

[0037] The result fusion module is used to fuse the post-processing result with the remote sensing image to be extracted to obtain the final extraction result.

[0038] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described remote sensing building extraction methods are implemented.

[0039] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described remote sensing building extraction methods.

[0040] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned remote sensing building extraction methods.

[0041] The present invention provides a remote sensing building extraction method and device, which obtains a remote sensing image to be extracted; inputs the remote sensing image to be extracted into an improved D-linknet model to obtain a corresponding extraction result; post-processes the extraction result to obtain a corresponding post-processing result; and fuses the post-processing result with the remote sensing image to be extracted to obtain a final extraction result. The present invention can improve the extraction accuracy by using the improved D-linknet model to extract buildings, and can remove isolated points by post-processing the extraction results, making the extraction results more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the present invention or 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 It is a schematic diagram of the flow of the remote sensing building extraction method provided by the present invention;

[0044] Figure 2 It is a schematic diagram of the improved D-linknet network structure provided by the present invention;

[0045] Figure 3 It is a schematic diagram of the structure of the central part of the original D-linknet provided by the present invention;

[0046] Figure 4 It is a structural schematic diagram of the remote sensing building extraction device provided by the present invention;

[0047] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Combine the following Figure 1 A remote sensing building extraction method of the present invention is described, comprising:

[0050] Step 100: Obtain the remote sensing image to be extracted;

[0051] Step 200: inputting the remote sensing image to be extracted into the improved D-linknet model to obtain the corresponding extraction result;

[0052] Specifically, the improved D-linknet model used in the embodiment of the present invention can better retain the detail information of the remote sensing image, so that the extracted remote sensing image has higher accuracy.

[0053] Step 300: post-processing the extraction result to obtain a corresponding post-processing result;

[0054] Specifically, after extracting the remote sensing image, the embodiment of the present invention further processes the extracted result to make the obtained result more accurate and more effective.

[0055] Step 400: Fusing the post-processing result with the remote sensing image to be extracted to obtain a final extraction result.

[0056] The present invention provides a remote sensing building extraction method, which includes obtaining a remote sensing image to be extracted; inputting the remote sensing image to be extracted into an improved D-linknet model to obtain a corresponding extraction result; post-processing the extraction result to obtain a corresponding post-processing result; and fusing the post-processing result with the remote sensing image to be extracted to obtain a final extraction result. The present invention can improve the extraction accuracy by using the improved D-linknet model to extract buildings, and can remove isolated points by post-processing the extraction results, making the extraction results more accurate.

[0057] Further, combined with Figure 2 As shown, according to the remote sensing building extraction method provided by the present invention, the improved D-linknet model includes a new encoding module, a new hole convolution module, a first channel attention module, a second channel attention module and a decoding module:

[0058] The new encoding module is obtained by adding at least one skip-layer connection between the layers in the encoding module of the original D-linknet model;

[0059] The new dilated convolution module is obtained by adjusting the dilated rate of the dilated convolution module of the original D-linknet model;

[0060] The first channel attention module is between the new encoding module and the new hole convolution module, and the second channel attention module is between the new hole convolution module and the decoding module.

[0061] Specifically, the original D-linknet network is improved in an embodiment of the present invention, and the specific improvement is as follows: the encoding module of the original D-linknet network includes five encoding layers, and the shallower convolutional layer features are added to the subsequent encoding layers by adding skip layer connections. For example, by adding two skip layer connections to the encoding layer, the output result of the first encoding layer and the output result of the second layer are mixed and input into the third layer, and then the mixed result is input into the fourth encoding layer. Adding these two skip connections first avoids the effect after multi-layer training being worse than the previous layer; in addition, the shallower convolutional layer features can be added through the skip connection. Those features have higher resolution and shallower layers, and will contain richer low-level information, which is more conducive to generating segmentation masks. Of course, other skip layer connection methods can also be used.

[0062] The central part of the original D-linknet is a dilated convolution using shortcut connections to form a serial-parallel structure with a dilation rate of 1, 2, 4, or 8. Figure 3 shown.

[0063] In order to avoid the problem of discontinuous dilated convolution, it is necessary to adjust the dilation rate of the central part of D-linknet. Referring to the HDC (Hybrid Dilated Convolution) structure, based on the goal of making the dilation rate have no common divisor greater than 1, this method adjusts the dilation rate of the central part to 1, 2, 5, 7, such as Figure 2 As shown in part b, the adjusted dilation rate allows each set of results before and after the convolution to be intertwined and interdependent, solving the problem of local information loss in dilated convolution, capturing multi-scale information, obtaining local information dependence, and solving the problem of inconsistent local information.

[0064] In the center of the original D-linknet, a channel attention module is added before and after the dilated convolution using shortcut connection, as shown in part c of 2. The Channel Attention module focuses on what features are meaningful. Its function is to assign different weights to each channel, so that the network can focus on important features and suppress unimportant features. When there are many channels at the bottom layer of the network, adding two Channel Attention modules helps to obtain more important features.

[0065] Further, according to the remote sensing building extraction method provided by the present invention, the improved D-linknet model parameters are trained by the following method:

[0066] Step 310: Obtain a remote sensing data set for training;

[0067] Step 320: preprocessing the remote sensing data set for training to obtain a preprocessed remote sensing data set;

[0068] Step 330: Divide the processed remote sensing data set into a training data set and a test data set;

[0069] Step 340: inputting the training data set into the improved D-linknet model for training until the corresponding loss function converges, thereby obtaining an improved D-linknet model with corresponding parameters;

[0070] Step 350: input the test data set into the improved D-linknet model of the corresponding parameters to obtain the corresponding accuracy;

[0071] Step 360: Compare the accuracy with a preset accuracy;

[0072] Step 370: if the accuracy is greater than the preset accuracy, re-execute step 340; if the accuracy is not greater than the preset accuracy, execute step 380;

[0073] Step 380: Stop training and fix the improved D-linknet model parameters.

[0074] Specifically, in the embodiment of the present invention, remote sensing image data labels are produced according to the acquired remote sensing image data, and then the remote sensing image data and the data labels are preprocessed to obtain preprocessing results. The preprocessing results are then randomly divided into a training data set and a test data set, and the data of the training data set is input into the network for training; data enhancement is used during the training process, and for each image, three enhancement methods are randomly selected from up and down flipping, left and right inversion, random angle rotation, random brightness adjustment, random contrast enhancement, and random sharpness enhancement for processing; training is performed until the loss converges. The test set data is placed in the trained improved D-linknet model network for building extraction; the index is calculated, and if the index meets the accuracy requirement, the trained improved D-linknet model network is used as the final model network. If not, the original D-linknet model is retrained according to the training data set.

[0075] Further, according to the remote sensing building extraction method provided by the present invention, the preprocessing of the remote sensing data set used for training to obtain the preprocessed remote sensing data set specifically includes:

[0076] Vectorizing the buildings in the remote sensing image in the remote sensing data set to obtain a vector layer;

[0077] Convert the vectorized layer into a raster layer to generate a label image;

[0078] The remote sensing image and the corresponding label image are cropped to the same size to obtain a processed remote sensing data set.

[0079] Specifically, open the remote sensing image in Arcmap; vectorize the buildings in the remote sensing image to obtain a vector layer; convert the vectorized layer to a raster layer to generate a label image. The processed remote sensing image and label are cropped to the same size, and the remote sensing image and label corresponding to the image content are named the same.

[0080] Further, according to the remote sensing building extraction method provided by the present invention, the step of inputting the training data set into the improved D-linknet model for training comprises:

[0081] During the training process, data enhancement is performed on each image in the training data set to obtain an enhanced image;

[0082] The improved D-linknet model is trained according to the enhanced image.

[0083] Specifically, data enhancement is used during the training process. For each image, three enhancement methods are randomly selected from up and down flipping, left and right inversion, random angle rotation, random brightness adjustment, random contrast enhancement, and random sharpness enhancement for processing.

[0084] Further, according to the remote sensing building extraction method provided by the present invention, the post-processing of the extraction result to obtain the corresponding post-processing result specifically includes:

[0085] Performing isolated point removal processing on the extraction result to obtain an extraction result after the isolated points are removed;

[0086] Rectangle optimization is performed on the extraction result after removing isolated points to obtain a corresponding post-processing result.

[0087] Specifically, first, the isolated points are removed from the obtained processing results to make the obtained processing results purer. Generally, the shape of a building is a rectangle, but it is often difficult to extract an accurate rectangle from the segmentation results of the improved D-linknet model. Therefore, a circumscribed rectangle is made for a connected domain of a certain area to optimize the extraction effect.

[0088] Combination Figure 4 The present invention provides a remote sensing building extraction device, comprising:

[0089] A remote sensing image acquisition module 41 is used to acquire the remote sensing image to be extracted;

[0090] A result extraction module 42, used for inputting the remote sensing image to be extracted into the improved D-linknet model to obtain a corresponding extraction result;

[0091] A post-processing module 43 is specifically used to post-process the extraction result to obtain a corresponding post-processing result;

[0092] The result fusion module 44 is used to fuse the post-processing result with the remote sensing image to be extracted to obtain the final extraction result.

[0093] Since the device provided in the embodiment of the present invention can be used to execute the method described in the above embodiment, its working principle and beneficial effects are similar, so they are not described in detail here. For specific contents, please refer to the introduction of the above embodiment.

[0094] The present invention provides a remote sensing building extraction device, which obtains a remote sensing image to be extracted; inputs the remote sensing image to be extracted into an improved D-linknet model to obtain a corresponding extraction result; post-processes the extraction result to obtain a corresponding post-processing result; and fuses the post-processing result with the remote sensing image to be extracted to obtain a final extraction result. The present invention can improve the extraction accuracy by using the improved D-linknet model to extract buildings, and can remove isolated points by post-processing the extraction results, making the extraction results more accurate.

[0095] Further, according to the remote sensing building extraction system provided by the present invention, the improved D-linknet model includes a new encoding module, a new hole convolution module, a first channel attention module, a second channel attention module and a decoding module:

[0096] The new encoding module is obtained by adding at least one skip-layer connection between the layers in the encoding module of the original D-linknet model;

[0097] The new dilated convolution module is obtained by adjusting the dilated rate of the dilated convolution module of the original D-linknet model;

[0098] The first channel attention module is between the new encoding module and the new hole convolution module, and the second channel attention module is between the new hole convolution module and the decoding module.

[0099] Further, according to the remote sensing building extraction system provided by the present invention, in the result extraction module 42, the improved D-linknet model parameters are trained by the following method:

[0100] Step 310: Obtain a remote sensing data set for training;

[0101] Step 320: preprocessing the remote sensing data set for training to obtain a preprocessed remote sensing data set;

[0102] Step 330: Divide the processed remote sensing data set into a training data set and a test data set;

[0103] Step 340: inputting the training data set into the improved D-linknet model for training until the corresponding loss function converges, thereby obtaining an improved D-linknet model with corresponding parameters;

[0104] Step 350: input the test data set into the improved D-linknet model of the corresponding parameters to obtain the corresponding accuracy;

[0105] Step 360: Compare the accuracy with a preset accuracy;

[0106] Step 370: if the accuracy is greater than the preset accuracy, re-execute step 340; if the accuracy is not greater than the preset accuracy, execute step 380;

[0107] Step 380: Stop training and fix the improved D-linknet model parameters.

[0108] Further, according to the remote sensing building extraction system provided by the present invention, in the result extraction module 42, the preprocessing of the remote sensing data set used for training to obtain the preprocessed remote sensing data set specifically includes:

[0109] Vectorizing the buildings in the remote sensing image in the remote sensing data set to obtain a vector layer;

[0110] Convert the vectorized layer into a raster layer to generate a label image;

[0111] The remote sensing image and the corresponding label image are cropped to the same size to obtain a processed remote sensing data set.

[0112] Further, according to the remote sensing building extraction method provided by the present invention, in the result extraction module 42, the step of inputting the training data set into the improved D-linknet model for training includes:

[0113] During the training process, data enhancement is performed on each image in the training data set to obtain an enhanced image;

[0114] The improved D-linknet model is trained according to the enhanced image.

[0115] Further, according to the remote sensing building extraction method provided by the present invention, in the result extraction module 42, the extraction result is post-processed to obtain the corresponding post-processing result, which specifically includes:

[0116] Performing isolated point removal processing on the extraction result to obtain an extraction result after the isolated points are removed;

[0117] Rectangle optimization is performed on the extraction result after removing isolated points to obtain a corresponding post-processing result.

[0118] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a remote sensing building extraction method, which includes: obtaining a remote sensing image to be extracted; inputting the remote sensing image to be extracted into the improved D-linknet model to obtain a corresponding extraction result; post-processing the extraction result to obtain a corresponding post-processing result; fusing the post-processing result with the remote sensing image to be extracted to obtain a final extraction result.

[0119] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0120] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a remote sensing building extraction method provided by the above methods, which method includes: obtaining a remote sensing image to be extracted; inputting the remote sensing image to be extracted into an improved D-linknet model to obtain a corresponding extraction result; post-processing the extraction result to obtain a corresponding post-processing result; fusing the post-processing result with the remote sensing image to be extracted to obtain a final extraction result.

[0121] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute a remote sensing building extraction method provided by the above methods, the method comprising: obtaining a remote sensing image to be extracted; inputting the remote sensing image to be extracted into an improved D-linknet model to obtain a corresponding extraction result; post-processing the extraction result to obtain a corresponding post-processing result; fusing the post-processing result with the remote sensing image to be extracted to obtain a final extraction result.

[0122] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may 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 labor.

[0123] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote sensing building extraction method, characterized in that: include: Obtaining remote sensing images to be extracted; Inputting the remote sensing image to be extracted into the improved D-linknet model to obtain the corresponding extraction result; Post-processing the extraction result to obtain a corresponding post-processing result; Fusing the post-processing result with the remote sensing image to be extracted to obtain a final extraction result; The improved D-linknet model includes a new encoding module, a new hole convolution module, a first channel attention module, a second channel attention module and a decoding module: The new encoding module is obtained by adding at least one skip-layer connection between the layers in the encoding module of the original D-linknet model; The new dilated convolution module is obtained by adjusting the dilated rate of the dilated convolution module of the original D-linknet model; The first channel attention module is between the new encoding module and the new hole convolution module, and the second channel attention module is between the new hole convolution module and the decoding module; The improved D-linknet model parameters are trained by the following method: Step 310: Obtain a remote sensing data set for training; Step 320: preprocessing the remote sensing data set for training to obtain a preprocessed remote sensing data set; Step 330: Divide the processed remote sensing data set into a training data set and a test data set; Step 340: inputting the training data set into the improved D-linknet model for training until the corresponding loss function converges, thereby obtaining an improved D-linknet model with corresponding parameters; Step 350: input the test data set into the improved D-linknet model of the corresponding parameters to obtain the corresponding accuracy; Step 360: Compare the accuracy with a preset accuracy; Step 370: if the accuracy is greater than the preset accuracy, re-execute step 340; if the accuracy is not greater than the preset accuracy, execute step 380; Step 380: Stop training and fix the improved D-linknet model parameters.

2. The remote sensing building extraction method according to claim 1, characterized in that: The preprocessing of the remote sensing data set for training to obtain the preprocessed remote sensing data set specifically includes: Vectorizing the buildings in the remote sensing image in the remote sensing dataset to obtain a vectorized layer; Convert the vectorized layer into a raster layer to generate a label image; The remote sensing image and the corresponding label image are cropped to the same size to obtain a processed remote sensing data set.

3. The remote sensing building extraction method according to claim 1, characterized in that: The step of inputting the training data set into the improved D-linknet model for training comprises: During the training process, data enhancement is performed on each image in the training data set to obtain an enhanced image; The improved D-linknet model is trained according to the enhanced image.

4. The remote sensing building extraction method according to claim 1, characterized in that: The post-processing of the extraction result to obtain the corresponding post-processing result specifically includes: Performing isolated point removal processing on the extraction result to obtain an extraction result after the isolated points are removed; The extraction result after removing isolated points is subjected to a minimum circumscribed rectangle optimization operation to obtain a corresponding post-processing result.

5. A remote sensing building extraction device, characterized in that: include: A remote sensing image acquisition module is used to acquire remote sensing images to be extracted; A result extraction module, used for inputting the remote sensing image to be extracted into the improved D-linknet model to obtain the corresponding extraction result; A post-processing module, specifically used to post-process the extraction result to obtain a corresponding post-processing result; A result fusion module is used to fuse the post-processing result with the remote sensing image to be extracted to obtain a final extraction result; The improved D-linknet model includes a new encoding module, a new hole convolution module, a first channel attention module, a second channel attention module and a decoding module: The new encoding module is obtained by adding at least one skip-layer connection between the layers in the encoding module of the original D-linknet model; The new dilated convolution module is obtained by adjusting the dilated rate of the dilated convolution module of the original D-linknet model; The first channel attention module is between the new encoding module and the new hole convolution module, and the second channel attention module is between the new hole convolution module and the decoding module; The improved D-linknet model parameters are trained by the following method: Step 310: Obtain a remote sensing data set for training; Step 320: preprocessing the remote sensing data set for training to obtain a preprocessed remote sensing data set; Step 330: Divide the processed remote sensing data set into a training data set and a test data set; Step 340: inputting the training data set into the improved D-linknet model for training until the corresponding loss function converges, thereby obtaining an improved D-linknet model with corresponding parameters; Step 350: input the test data set into the improved D-linknet model of the corresponding parameters to obtain the corresponding accuracy; Step 360: Compare the accuracy with a preset accuracy; Step 370: if the accuracy is greater than the preset accuracy, re-execute step 340; if the accuracy is not greater than the preset accuracy, execute step 380; Step 380: Stop training and fix the improved D-linknet model parameters.

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, the steps of the remote sensing building extraction method according to any one of claims 1 to 4 are implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the remote sensing building extraction method according to any one of claims 1 to 4 are implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the remote sensing building extraction method according to any one of claims 1 to 4 are implemented.

Images