A Remote Sensing Image Building Change Detection Method and System

Abstract

The present invention provides a method and system for detecting building changes in remote sensing images, belonging to the technical field of remote sensing detection, including: inputting dual-temporal optical images into a cyclic alignment Siamese network between remote sensing images, and inputting the original image, the first corrected image, and the second corrected image into a multi-task bidirectional feature fusion framework respectively to obtain multiple sets of building detection results; using a multi-task collaborative optimization loss function to adjust the multiple sets of building detection results to obtain a building change detection model for remote sensing images; inputting the multi-level features of the dual-temporal optical images of the building to be detected into the building change detection model for remote sensing images, and outputting the building change detection results for remote sensing images. The present invention overcomes the problem of performance degradation of deep learning networks caused by parallax commonly existing in the research of high-resolution remote sensing image building change detection technology, and effectively improves the performance of high-resolution remote sensing image building change detection under the influence of obvious parallax.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing detection, and particularly relates to a method and system for detecting building changes in remote sensing images. Background Art

[0002] Change detection is one of the important research contents in the field of remote sensing vision applications. It aims at remote sensing image data observed at different times in the same area, and uses image processing technology to extract image features of different time phases, so as to locate the changed areas and unchanged areas on the images. Since the surface ecosystem and human social activities are both dynamic and constantly evolving, change detection has been widely applied in various application fields, such as land cover change, disaster assessment, urban planning and expansion, etc. In recent years, with the continuous development of earth observation technology, a large amount of hyperspectral, high spatio-temporal resolution remote sensing data has emerged continuously. Remote sensing data has the characteristics of short revisit period, wide coverage, high spatial resolution and wide spectral resolution. Therefore, multi-temporal high-resolution remote sensing image data has become the main data source for change detection tasks.

[0003] Change detection based on twin fully convolutional neural networks is often a pixel-level detection task. It adopts an Encoder-Decoder architecture. This architecture first uses a convolutional neural network encoder with shared weights to extract low-level and high-level features of the bi-temporal images respectively. Then, it uses a decoder to obtain the difference features and gradually restore the resolution along the bottom-up path. Finally, a classifier is connected to predict the changed areas. Therefore, whether the features extracted by the convolutional neural network are aligned is an important factor affecting the accuracy of change detection. In the change detection task, the main reason affecting feature alignment is the feature misalignment phenomenon caused by the geometric registration error of bi-temporal images. Image registration refers to finding the spatial mapping relationship between the pixels of one image and the corresponding pixels of another image. The image pairs can be multi-temporal images, multi-sensor images, etc. The spatial mapping relationship can be translation, rotation, homography transformation, or more complex deformation models. Change detection refers to obtaining the changed areas through multi-source remote sensing images taken at different times in the same area. Therefore, the existing change detection methods require high-precision image registration of the multi-temporal remote sensing images involved. Image registration error is one of the main error sources in change detection. If high-precision image registration cannot be performed, a large number of misdetection areas will be predicted. Feature-based image registration methods include three main steps: key point detection and feature description, feature matching, and image transformation, etc. Generally, high registration accuracy can be achieved for most ground points in image registration. However, for buildings above and below the ground, there are often large errors in image registration. This is because the imaging method or photographing method of optical remote sensing images often uses central projection. The electromagnetic waves of the ground objects received by the sensor are not completely perpendicular to the ground shooting area. When the ground fluctuates, the image points of the ground points above or below the base surface will shift on the imaging plane, which is called projection error. For example, for remote sensing images taken from the same perspective, the higher the building, the greater its projection error on the image. When the same building is photographed from different perspectives or photographing points, its projection error on the image is also different. When the shooting perspective changes significantly, the offset amplitude of the top area of the same high-rise building between the two images may even exceed the scale of a single building. This deviation caused by the visual difference in image shooting, where the image points of the points above the ground in the scene do not coincide in the bi-temporal remote sensing images, is called parallax. Therefore, on the bi-temporal remote sensing images taken from the same perspective in the same area, the projection error of the building changes consistently with the height, and the height of the building has no effect on the change detection result. However, when the shooting perspectives of the bi-temporal remote sensing images are different, the parallax will cause the same building to shift in the two images, and this offset problem caused by the parallax still exists even after image registration.

[0004] Therefore, a method for building change detection in high-resolution remote sensing images based on a parallax-aware depth network needs further exploration and improvement. Summary of the Invention

[0005] The present invention provides a method and system for detecting building changes in remote sensing images, aiming to solve the defect that the performance of the deep network degrades in the existing building change detection technology when there is an obvious parallax between two-phase optical images in the prior art.

[0006] In a first aspect, the present invention provides a method for detecting building changes in remote sensing images, including:

[0007] Collecting dual-phase optical images of the original building;

[0008] Inputting the dual-phase optical images into a cyclic alignment Siamese network between remote sensing images to obtain an original image, a primary corrected image, and a secondary corrected image;

[0009] Inputting the original image, the primary corrected image, and the secondary corrected image into a multi-task bidirectional feature fusion framework respectively to obtain multiple groups of building detection results;

[0010] Using a multi-task collaborative optimization loss function to adjust the multiple groups of building detection results to obtain a building change detection model for remote sensing images;

[0011] Inputting the multi-level features of the dual-phase optical images of the building to be detected into the building change detection model for remote sensing images, and outputting the building change detection result for remote sensing images.

[0012] According to the method for detecting building changes in remote sensing images provided by the present invention, the dual-phase optical images include a first-phase optical image and a second-phase optical image.

[0013] According to the method for detecting building changes in remote sensing images provided by the present invention, inputting the multi-level features of the dual-phase optical images into a cyclic alignment Siamese network between remote sensing images to obtain an original image, a primary corrected image, and a secondary corrected image, including:

[0014] Determining that the cyclic alignment Siamese network between remote sensing images includes a symmetric optical flow generation network and a cyclic alignment module, and the symmetric optical flow generation network includes a first encoder, a feature alignment module, and a first decoder;

[0015] Inputting the first-phase optical image and the second-phase optical image into the first encoder to obtain multi-level features of the first-phase optical image and multi-level features of the second-phase optical image;

[0016] Inputting the multi-level features of the first-phase optical image and the multi-level features of the second-phase optical image into the feature alignment module to obtain multi-level features with the influence of the inter-phase parallax eliminated;

[0017] Input the multi-level features that eliminate the influence of temporal parallax into the first decoder to obtain the first optical flow and the second optical flow;

[0018] Input the first optical flow and the second optical flow into the cyclic alignment module to obtain the original image, the first corrected image, and the second corrected image.

[0019] According to a method for detecting building changes in remote sensing images provided by the present invention, input the original image, the first corrected image, and the second corrected image into a multi-task bidirectional feature fusion framework respectively to obtain multiple groups of building detection results, including:

[0020] Determine that the multi-task bidirectional feature fusion framework includes a second encoder, a second decoder, and a bidirectional feature fusion module;

[0021] Input the original image, the first corrected image, and the second corrected image into the second encoder respectively to obtain multiple groups of multi-level features;

[0022] Input each group of multi-level features into the second decoder respectively to obtain multiple groups of preset high-level multi-level features;

[0023] Input each group of preset high-level multi-level features into the bidirectional feature fusion module to obtain the multiple groups of building detection results.

[0024] According to a method for detecting building changes in remote sensing images provided by the present invention, the first encoder and the second encoder share weights.

[0025] According to a method for detecting building changes in remote sensing images provided by the present invention, use a multi-task collaborative optimization loss function to adjust the multiple groups of building detection results to obtain a remote sensing image building change detection model.

[0026] L = L c + λ1L b + λ2L′ b + λ3L″ b

[0027]

[0028]

[0029]

[0030]

[0031] Among them, L represents the total loss, L c represents the change detection loss, L b , L′ b , L″ brespectively represent the original detection loss of the building, the first detection loss of the building, and the second detection loss of the building. λ1, λ2, and λ3 are the weight coefficients of L b , L′ b , L″ b . N represents the total number of pixels in each image, i represents the direction of the change detection result. i = 1 represents the first optical flow t1 to the second optical flow t2, i = 2 represents the second optical flow t2 to the first optical flow t1, n represents the pixel serial number, and z n represents the change detection label, represents the change detection result, respectively represent the original building label and the corrected building label, represents the building detection result, and α, γ represent the adjustment setting parameters in the building detection loss.

[0032] In a second aspect, the present invention further provides a remote sensing image building change detection system, including:

[0033] An acquisition module for acquiring dual-temporal optical images of the original building;

[0034] A first processing module for inputting the dual-temporal optical images into a remote sensing image inter-cycle alignment siamese network to obtain an original image, a first corrected image, and a second corrected image;

[0035] A second processing module for inputting the original image, the first corrected image, and the second corrected image into a multi-task bidirectional feature fusion framework respectively to obtain multiple groups of building detection results;

[0036] A third processing module for using a multi-task collaborative optimization loss function to adjust the multiple groups of building detection results to obtain a remote sensing image building change detection model;

[0037] A detection module for inputting the multi-level features of the dual-temporal optical images of the building to be detected into the remote sensing image building change detection model and outputting a remote sensing image building change detection result.

[0038] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the remote sensing image building change detection method as described in any one of the above.

[0039] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the remote sensing image building change detection method as described in any one of the above.

[0040] Fifth aspect, the present invention also provides a computer program product, including a computer program which, when executed by a processor, implements the remote sensing image building change detection method as described in any one of the above.

[0041] The remote sensing image building change detection method and system provided by the present invention overcome the problem of performance degradation of deep learning networks caused by parallax commonly existing in the research of high-resolution remote sensing image building change detection technology by applying a multi-temporal cyclic alignment siamese network based on parallax perception, and effectively improve the performance of high-resolution remote sensing image building change detection under the influence of obvious parallax. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0043] Figure 1 is one of the schematic flowcharts of the remote sensing image building change detection method provided by the present invention;

[0044] Figure 2 is the second schematic flowchart of the remote sensing image building change detection method provided by the present invention;

[0045] Figure 3 is the structural diagram of the cyclic alignment siamese network provided by the present invention;

[0046] Figure 4 is the schematic diagram of the optical flow correction principle provided by the present invention;

[0047] Figure 5 is the multi-task bidirectional feature fusion framework diagram provided by the present invention;

[0048] Figure 6 is the structural diagram of the remote sensing image building change detection system provided by the present invention;

[0049] Figure 7 is the structural diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts fall within the protection scope of the present invention.

[0051] Figure 1 is one of the schematic flowcharts of the remote sensing image building change detection method provided by the embodiments of the present invention. As Figure 1 shown, it includes:

[0052] Step 100: Collect dual-temporal optical images of the original building;

[0053] Step 200: Input the dual-temporal optical images into the cyclic alignment Siamese network between remote sensing images to obtain the original image, the first corrected image, and the second corrected image;

[0054] Step 300: Input the original image, the first corrected image, and the second corrected image into the multi-task bidirectional feature fusion framework respectively to obtain multiple sets of building detection results;

[0055] Step 400: Use the multi-task collaborative optimization loss function to adjust the multiple sets of building detection results to obtain the remote sensing image building change detection model;

[0056] Step 500: Input the multi-level features of the dual-temporal optical images of the building to be detected into the remote sensing image building change detection model, and output the remote sensing image building change detection result.

[0057] Specifically, the high-resolution remote sensing image building change detection method based on the disparity perception depth network proposed by the embodiments of the present invention mainly includes a dual-temporal cyclic alignment Siamese network, a multi-task bidirectional feature fusion framework, and a multi-task collaborative building change detection loss function.

[0058] The overall process is shown in Figure 2 As shown. First, collect the dual-temporal optical images of the original building, including I1 and I2, and process them through the designed cyclic alignment Siamese network structure between two-temporal high-resolution remote sensing images. Pass through the encoder (Encoder) and the flow decoder in sequence, and obtain the first corrected images I′1 and I′2, as well as the second optical flow twist (Reconstruction with flow 12 and Wrap withflow 21 ) through the first optical flow twist (Wrap with flow 12and Reconstruction with flow 21 ) Obtain the secondarily corrected images I″1 and I″2; then input I1 and I2, I′1 and I′2, I″1 and I″2 into the multi - task bidirectional feature fusion framework, which also goes through an encoder (Encoder) and a decoder (Buildingdecoder) to obtain multiple sets of building detection results, namely M1 and M2, M′1 and M′2, M″1 and M″2. Combine the building change detection loss function with multi - task collaboration to obtain a remote sensing image building change detection model, and output the remote sensing image building change detection results C1 and C2.

[0059] The present invention overcomes the problem of performance degradation of deep learning networks caused by parallax commonly existing in the research of high - resolution remote sensing image building change detection technology by applying a multi - temporal cyclic alignment siamese network based on parallax perception, and effectively improves the performance of high - resolution remote sensing image building change detection under the influence of obvious parallax.

[0060] Based on the above - mentioned embodiment, the two - temporal optical images include a first - temporal optical image and a second - temporal optical image.

[0061] Input the multi - level features of the two - temporal optical images into the cyclic alignment siamese network between remote sensing images to obtain the original image, the first - corrected image, and the second - corrected image, including:

[0062] Determine that the cyclic alignment siamese network between remote sensing images includes a symmetric optical flow generation network and a cyclic alignment module, and the symmetric optical flow generation network includes a first encoder, a feature alignment module, and a first decoder;

[0063] Input the first - temporal optical image and the second - temporal optical image into the first encoder to obtain the multi - level features of the first - temporal optical image and the multi - level features of the second - temporal optical image;

[0064] Input the multi - level features of the first - temporal optical image and the multi - level features of the second - temporal optical image into the feature alignment module to obtain multi - level features that eliminate the influence of inter - temporal parallax;

[0065] Input the multi - level features that eliminate the influence of inter - temporal parallax into the first decoder to obtain the first optical flow and the second optical flow;

[0066] Input the first optical flow and the second optical flow into the cyclic alignment module to obtain the original image, the first - corrected image, and the second - corrected image.

[0067] Specifically, as Figure 3As shown in the figure, the cyclic alignment Siamese network consists of a symmetric optical flow generation network and a cyclic alignment module, which generate the correction results of two-temporal images. The symmetric optical flow generation network consists of three parts: an encoder, a feature alignment module, and a decoder. Taking paired two-temporal optical image data as input, the encoder extracts multi-level features of the two-temporal optical images respectively. The multi-level features (Multi-level Features) are input into the Feature AlignModule, where the influence of inter-temporal parallax is eliminated at the feature level. The multi-level features after eliminating the parallax influence are input into the decoder to obtain optical flows in two directions. In the cyclic alignment module, the optical flows in two directions are used to correct the two-temporal optical images. Each temporal optical image is corrected twice by the two optical flows. Finally, two original images, two once-corrected images, and two twice-corrected images can be obtained, where DFA represents different feature alignment operations, SFA represents multi-scale feature alignment operations, and C represents Concatenation connection operations.

[0068] The optical flow-based image correction principle therein is as Figure 4 shown. Taking any feature T1 in the first-temporal optical image and any feature T2 in the second-temporal optical image as examples respectively, optical flow image correction and alignment operations are performed.

[0069] Through the cyclic alignment Siamese network, the present invention obtains optical flow without optical flow labels, and the obtained optical flow is used to alleviate the misalignment problem between double-temporal remote sensing images caused by the parallax problem at the image level.

[0070] Based on the above embodiments, the original image, the once-corrected image, and the twice-corrected image are respectively input into the multi-task bidirectional feature fusion framework to obtain multiple groups of building detection results, including:

[0071] It is determined that the multi-task bidirectional feature fusion framework includes a second encoder, a second decoder, and a bidirectional feature fusion module;

[0072] The original image, the once-corrected image, and the twice-corrected image are respectively input into the second encoder to obtain multiple groups of multi-level features;

[0073] Each group of multi-level features is respectively input into the second decoder to obtain multiple groups of preset high-level multi-level features;

[0074] Each group of preset high-level multi-level features is input into the bidirectional feature fusion module to obtain the multiple groups of building detection results.

[0075] Among them, the first encoder and the second encoder share weights.

[0076] Specifically, as Figure 5The multi-task bidirectional feature fusion framework shown, which is composed of an encoder, a decoder, and a bidirectional feature fusion module (BFAM). Among them, the encoder shares weights with the encoder in the foregoing embodiment. The six images obtained in the foregoing embodiment are respectively input into the encoder, and a set of multi-level features can be obtained for each image. Each set of multi-level features is respectively input into the decoder to obtain a set of high-level multi-level features. The high-level multi-level features are input into the bidirectional feature fusion framework to obtain two building change detection results. At the same time, after passing through the convolutional layer, the high-level multi-level features can obtain six sets of building detection results.

[0077] The method of the present invention is based on a multi-task bidirectional feature fusion framework, which realizes the organic combination between the building detection task and the change detection task, and fuses the bidirectional features obtained by the multi-task framework, further improving the accuracy of building change detection.

[0078] Based on the above embodiment, using a multi-task collaborative optimization loss function to adjust the multi-group building detection results to obtain a remote sensing image building change detection model, including:

[0079] L = L c + λ1L b + λ2L′ b + λ3L″ b

[0080]

[0081]

[0082]

[0083]

[0084] Among them, L represents the total loss, L c represents the change detection loss, L b , L′ b , L″ b respectively represent the original building detection loss, the first building detection loss, and the second building detection loss, and λ1, λ2, and λ3 are the weight coefficients of L b , L′ b , L″ b respectively, N represents the total number of pixels in each image, i represents the direction of the change detection result, i = 1 represents the first optical flow t1 to the second optical flow t2, i = 2 represents the second optical flow t2 to the first optical flow t1, n represents the pixel number, z n represents the change detection label, represents the change detection result, respectively represent the original building label and the corrected building label, Indicates the building detection result, and α, γ represent the adjustment setting parameters in the building detection loss.

[0085] The method for detecting building changes in high-resolution remote sensing images based on a disparity-aware depth network of the present invention not only utilizes the powerful feature extraction and reasoning capabilities of deep learning methods, but also combines the design of a special multi-task cyclic alignment network structure. The adopted cyclic alignment network structure can effectively eliminate the disparity impact between two-phase optical images and provide more constraint conditions for the training of the depth network, improving the stability and robustness of the training; the multi-task bidirectional feature fusion structure realizes the organic combination between the building detection task and the change detection task, and fuses the bidirectional features obtained by the multi-task framework, further improving the accuracy of building change detection.

[0086] The remote sensing image building change detection system provided by the present invention will be described below. The remote sensing image building change detection system described below can be correspondingly referred to the remote sensing image building change detection method described above.

[0087] Figure 6 is a schematic structural diagram of the remote sensing image building change detection system provided by an embodiment of the present invention, as Figure 6 shown, including: an acquisition module 61, a first processing module 62, a second processing module 63, a third processing module 64, and a detection module 65, wherein:

[0088] The acquisition module 61 is used to acquire two-phase optical images of the original building; the first processing module 62 is used to input the two-phase optical images into the cyclic alignment Siamese network between remote sensing images to obtain the original image, the first corrected image, and the second corrected image; the second processing module 63 is used to input the original image, the first corrected image, and the second corrected image into the multi-task bidirectional feature fusion framework respectively to obtain multiple sets of building detection results; the third processing module 64 is used to adjust the multiple sets of building detection results by using the multi-task collaborative optimization loss function to obtain the remote sensing image building change detection model; the detection module 65 is used to input the multi-level features of the two-phase optical images of the building to be detected into the remote sensing image building change detection model and output the remote sensing image building change detection result.

[0089] Figure 7 illustrates a schematic structural diagram of an electronic device, as Figure 7As shown in the figure, the electronic device may include: a processor 710, a communications interface 720, a memory 730, and a communication bus 740. Among them, the processor 710, the communications interface 720, and the memory 730 complete communication with each other through the communication bus 740. The processor 710 may call the logical instructions in the memory 730 to execute a method for detecting building changes in remote sensing images. The method includes: collecting dual-temporal optical images of the original building; inputting the dual-temporal optical images into a cyclic alignment siamese network between remote sensing images to obtain an original image, a first corrected image, and a second corrected image; respectively inputting the original image, the first corrected image, and the second corrected image into a multi-task bidirectional feature fusion framework to obtain multiple sets of building detection results; using a multi-task collaborative optimization loss function to adjust the multiple sets of building detection results to obtain a model for detecting building changes in remote sensing images; inputting multi-level features of the dual-temporal optical images of the building to be detected into the model for detecting building changes in remote sensing images, and outputting the detection results of building changes in remote sensing images.

[0090] In addition, when the logical instructions in the above-mentioned memory 730 are implemented in the form of software functional units and sold or used as an independent product, they may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc that can store program codes.

[0091] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the remote sensing image building change detection method provided by the above-mentioned various methods. The method includes: collecting dual-temporal optical images of the original building; inputting the dual-temporal optical images into a cyclic alignment siamese network between remote sensing images to obtain an original image, a first corrected image, and a second corrected image; respectively inputting the original image, the first corrected image, and the second corrected image into a multi-task bidirectional feature fusion framework to obtain multiple sets of building detection results; using a multi-task collaborative optimization loss function to adjust the multiple sets of building detection results to obtain a remote sensing image building change detection model; inputting the multi-level features of the dual-temporal optical images of the building to be detected into the remote sensing image building change detection model, and outputting the remote sensing image building change detection result.

[0092] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes the remote sensing image building change detection method provided by the above-mentioned various methods. The method includes: collecting dual-temporal optical images of the original building; inputting the dual-temporal optical images into a cyclic alignment siamese network between remote sensing images to obtain an original image, a first corrected image, and a second corrected image; respectively inputting the original image, the first corrected image, and the second corrected image into a multi-task bidirectional feature fusion framework to obtain multiple sets of building detection results; using a multi-task collaborative optimization loss function to adjust the multiple sets of building detection results to obtain a remote sensing image building change detection model; inputting the multi-level features of the dual-temporal optical images of the building to be detected into the remote sensing image building change detection model, and outputting the remote sensing image building change detection result.

[0093] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0094] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment 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 such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting building changes in remote sensing images, characterized in that, Including: Collecting dual-temporal optical images of the original building; Inputting the dual-temporal optical images into the cyclic alignment siamese network between remote sensing images to obtain the original image, the first corrected image, and the second corrected image; Respectively inputting the original image, the first corrected image, and the second corrected image into the multi-task bidirectional feature fusion framework to obtain multiple sets of building detection results; Using the multi-task collaborative optimization loss function to adjust the multiple sets of building detection results to obtain the remote sensing image building change detection model; Inputting the multi-level features of the dual-temporal optical images of the building to be detected into the remote sensing image building change detection model and outputting the remote sensing image building change detection results; Inputting the multi-level features of the dual-temporal optical images into the cyclic alignment siamese network between remote sensing images to obtain the original image, the first corrected image, and the second corrected image, including: Determining that the cyclic alignment siamese network between remote sensing images includes a symmetric optical flow generation network and a cyclic alignment module, and the symmetric optical flow generation network includes a first encoder, a feature alignment module, and a first decoder; Inputting the first-temporal optical image and the second-temporal optical image into the first encoder to obtain the multi-level features of the first-temporal optical image and the multi-level features of the second-temporal optical image; Inputting the multi-level features of the first-temporal optical image and the multi-level features of the second-temporal optical image into the feature alignment module to obtain the multi-level features that eliminate the influence of the inter-temporal parallax; Inputting the multi-level features that eliminate the influence of the inter-temporal parallax into the first decoder to obtain the first optical flow and the second optical flow; Inputting the first optical flow and the second optical flow into the cyclic alignment module to obtain the original image, the first corrected image, and the second corrected image; Respectively inputting the original image, the first corrected image, and the second corrected image into the multi-task bidirectional feature fusion framework to obtain multiple sets of building detection results, including: Determining that the multi-task bidirectional feature fusion framework includes a second encoder, a second decoder, and a bidirectional feature fusion module; Respectively inputting the original image, the first corrected image, and the second corrected image into the second encoder to obtain multiple sets of multi-level features; Respectively inputting each set of multi-level features into the second decoder to obtain multiple sets of preset high-level multi-level features; Inputting each set of preset high-level multi-level features into the bidirectional feature fusion module to obtain the multiple sets of building detection results.

2. The remote sensing image building change detection method according to claim 1, characterized in that The dual-temporal optical images include the first-temporal optical image and the second-temporal optical image.

3. The remote sensing image building change detection method according to claim 1, characterized in that The first encoder and the second encoder share weights.

4. The remote sensing image building change detection method according to claim 1, characterized in that Using the multi-task collaborative optimization loss function to adjust the multiple sets of building detection results to obtain the remote sensing image building change detection model, including: L = L c + λ1L b + λ2L′ b + λ3L″ b Among them, \(L\) represents the total loss, \(L\) c represents the change detection loss, \(L\) b , \(L'\) b , \(L''\) b respectively represent the original building detection loss, the first building detection loss, and the second building detection loss. \(\lambda_1\), \(\lambda_2\), and \(\lambda_3\) are the weight coefficients of \(L\) b , \(L'\) b , \(L''\) b respectively. \(N\) represents the total number of pixels in each image, \(i\) represents the direction of the change detection result. \(i = 1\) represents the first optical flow \(t_1\) to the second optical flow \(t_2\), \(i = 2\) represents the second optical flow \(t_2\) to the first optical flow \(t_1\), \(n\) represents the pixel number, \(z\) n represents the change detection label, represents the change detection result, respectively represent the original building label and the corrected building label, represents the building detection result. \(\alpha\), \(\gamma\) represent the adjustment setting parameters in the building detection loss.

5. A remote sensing image building change detection system, based on the remote sensing image building change detection method according to any one of claims 1 to 4, characterized in that, Including: A collection module for collecting dual-temporal optical images of the original building; A first processing module for inputting the dual-temporal optical images into the cyclic alignment siamese network between remote sensing images to obtain the original image, the first corrected image, and the second corrected image; A second processing module for respectively inputting the original image, the first corrected image, and the second corrected image into the multi-task bidirectional feature fusion framework to obtain multiple sets of building detection results; The third processing module is used to adjust the multiple groups of building detection results by using multi-task collaborative optimization of the loss function to obtain a remote sensing image building change detection model; The detection module is used to input the multi-level features of the bi-temporal optical images of the building to be detected into the remote sensing image building change detection model and output the remote sensing image building change detection result.

6. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, When the processor executes the program, it implements the remote sensing image building change detection method according to any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the remote sensing image building change detection method according to any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the remote sensing image building change detection method according to any one of claims 1 to 4.

Images