A remote sensing image change detection method fusing D-InSAR deformation supervision

Abstract

The application provides a remote sensing image change detection method fusing D-InSAR deformation supervision, and relates to the technical field of remote sensing information processing and artificial intelligence. The method comprises the following steps: acquiring double-time optical remote sensing images of a target area and corresponding SAR images; if the SAR images can be acquired, generating D-InSAR deformation information reflecting the ground deformation; pre-processing the optical remote sensing images and the D-InSAR deformation information; constructing a joint learning network composed of a shared encoder, a change detection decoder and a deformation estimation decoder; outputting four types of change detection probability maps and predicted deformation results through the joint learning network; in the joint learning network training stage, according to the D-InSAR deformation information acquisition state, it is selected whether to join the deformation supervision; in the inference stage, an adaptive deformation correction strategy is adopted; according to the corrected four types of change detection probability maps or the change detection decoder output result, the final change detection result is obtained; the accuracy and reliability of change detection in disaster scenarios can be improved.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing information processing and artificial intelligence technology, and in particular to a method for detecting changes in remote sensing images that integrates D-InSAR deformation supervision. Background Technology

[0002] Remote sensing change detection is a technology that identifies whether changes have occurred on the Earth's surface by comparing remote sensing images of the same area at different times. This technology is widely used in scenarios such as disaster emergency monitoring, urban construction management, building change identification, infrastructure damage assessment, and surface deformation analysis.

[0003] Change detection plays a crucial role in extreme weather disasters. For example, extreme rainfall can trigger floods, landslides, ground subsidence, and road damage; typhoons can cause building damage, shoreline changes, farmland damage, and infrastructure destruction. After a disaster, quickly and accurately identifying the areas and extent of change is of great significance for disaster assessment, emergency response, and post-disaster evaluation.

[0004] Existing change detection methods mostly rely on pre- and post-change optical remote sensing images, using deep learning models to extract the differences between the two images and generate change detection results. Optical remote sensing images have advantages such as clear texture and rich spatial details, making them suitable for identifying changes in land features such as buildings, roads, water bodies, and vegetation.

[0005] However, optical remote sensing images are also susceptible to changes in lighting, shadows, clouds, fog, seasonal differences, vegetation phenology, and imaging conditions. In disaster scenarios such as extreme rainfall and typhoons, the lighting and surface cover of pre- and post-disaster images often differ significantly. Models may misinterpret changes in shadow, humidity, and vegetation as real changes, thus producing false changes.

[0006] D-InSAR technology can acquire surface deformation information using SAR imagery, reflecting the true deformation of the land surface. Compared with relying solely on optical imagery, D-InSAR deformation information can provide a more reliable physical basis for change detection, and is especially suitable for identifying landslides, subsidence, building damage, and disaster-affected areas.

[0007] Currently, D-InSAR results are typically used as standalone deformation analysis products, with limited integration with optical remote sensing image change detection models. Existing methods also lack a flexible mechanism to selectively incorporate D-InSAR deformation information during the training or inference phases, depending on the availability of SAR data.

[0008] Therefore, it is necessary to propose a remote sensing image change detection method that integrates D-InSAR deformation supervision, so that the model can not only use optical images to identify apparent changes, but also use D-InSAR deformation information to determine whether the changes have actually occurred, and further classify the change results into four categories: no obvious change, slight change, moderate change, and significant change, thereby improving the accuracy and reliability of change detection in extreme meteorological disaster scenarios. Summary of the Invention

[0009] To address the technical problems of existing technologies, such as the susceptibility of optical images to interference leading to spurious changes and inaccurate detection results, and the failure to fully utilize deformation information to improve detection accuracy by deeply integrating D-InSAR deformation information with optical image change detection, this invention provides a remote sensing image change detection method that integrates D-InSAR deformation supervision. The technical solution is as follows: On the one hand, a remote sensing image change detection method integrating D-InSAR deformation supervision is provided, including: S1, acquire dual-temporal optical remote sensing images of the target area and corresponding SAR images of the area; the dual-temporal optical remote sensing images are optical remote sensing images before and after the change. S2, if SAR images can be acquired, D-InSAR deformation information reflecting surface deformation can be generated; S3 preprocesses optical remote sensing images and D-InSAR deformation information to ensure that the two are consistent in spatial location, resolution and image range. S4, construct a joint learning network consisting of a shared encoder, a change detection decoder, and a deformation estimation decoder; S5: Input the dual-temporal optical remote sensing images into the shared encoder to extract change features; S6. Input the change features into the change detection decoder to obtain four types of change detection probability maps: no obvious change, slight change, moderate change, and significant change. S7, input the change features into the deformation estimation decoder to obtain the predicted deformation result; S8. During the training phase, based on the acquisition status of D-InSAR deformation information, select whether to add deformation supervision: when D-InSAR deformation information is available, use the change ground truth label and D-InSAR deformation information to train the joint learning network; when D-InSAR deformation information is not available, use the change ground truth label to train the change detection branch. S9 employs an adaptive deformation correction strategy during the inference phase: when D-InSAR deformation information is available, it is used to correct the probability maps of the four types of changes; when D-InSAR deformation information is unavailable, the predicted deformation results or the output results of the change detection decoder are used to generate the change detection map. S10: Based on the corrected four-type change detection probability map or the output of the change detection decoder, the final change detection result is obtained.

[0010] Furthermore, the optical remote sensing imagery is used to provide information on the surface texture, color, and spatial structure of the target area before and after the changes. The preprocessing includes image registration, cropping, resampling, and normalization.

[0011] Furthermore, the D-InSAR deformation information is vertical deformation information, expressed as: ; in, Represents vertical deformation. This indicates the deformation of the radar line-of-sight direction. Indicates the radar incident angle.

[0012] Furthermore, the D-InSAR deformation information is vertical deformation information; The following observation equation is established: ; ; Solving for the vertical deformation component using least squares. and east-west deformation components , To cause deformation of the line of sight during ascent, To reduce the deformation along the line of sight, , These are the incident angles for the ascending and descending orbits, respectively. , These are the azimuth angles corresponding to the ascending and descending orbits, respectively.

[0013] Furthermore, the shared encoder is used to extract the change features of the optical remote sensing images before and after the change; Let the optical remote sensing image before the change be... The changed optical remote sensing image is The shared encoder is E, and the extracted variation features are: ; in, This represents the change features extracted from optical remote sensing images before and after the change.

[0014] Furthermore, the change detection decoder is used to output the probability of each pixel belonging to no significant change, slight change, moderate change, and significant change, respectively; Let the change detection decoder be The resulting probability maps for detecting four types of changes are as follows: ; in, Indicates the first The pixel belongs to the first Predicted probability of class change =0,1,2,3 This represents the change features extracted from the optical remote sensing images before and after the change, and Softmax represents the normalized exponential function.

[0015] Furthermore, the deformation estimation decoder is used to estimate the degree of deformation at each pixel position in the target region, that is, to output the predicted deformation result; Let the deformation estimation decoder be The predicted deformation results are as follows: ; in, Indicates the first The predicted deformation value of each pixel. This represents the change features extracted from optical remote sensing images before and after the change.

[0016] Furthermore, during the training phase, when D-InSAR deformation information is available, the joint learning network is trained using the change ground truth label and D-InSAR deformation information, and optimized using a joint loss function that includes change detection loss and deformation estimation loss; wherein, the change detection loss is used to constrain the four types of change detection results, and the deformation estimation loss is used to constrain the predicted deformation results. set up Indicates the first The pixel in the first truth labels of changes on the class Indicates the first The pixel belongs to the first Predicted probability of class change Indicates the number of pixels. To prevent abnormal logarithmic calculations or extremely small constants with a denominator of zero, the four-class cross-entropy loss is used. for: ; loss The degree of overlap between the predicted result and the true label is used to measure the degree of overlap, and is expressed as: ; Change detection loss Represented as: ; in, Indicates loss Weighting coefficients; set up Indicates the first D-InSAR deformation value of each pixel Indicates the first The predicted deformation value of each pixel. Indicates the first Does each pixel possess valid D-InSAR deformation information? If valid... =1, invalid =0, then the deformation estimation loss for: ; The joint loss function is expressed as: ; in, Indicates loss The weighting coefficients.

[0017] Furthermore, during the inference phase, when D-InSAR deformation information is available, the probability map for the four types of change detection is corrected using the D-InSAR deformation information; when D-InSAR deformation information is unavailable, but the predicted deformation result is available, the predicted deformation result is used for correction; when neither is available, a change detection map is generated based on the output result of the change detection decoder. Let the first The degree of deformation of each pixel is Then we have: ; in, =1 indicates that D-InSAR vertical deformation information is used. =0 indicates that the predicted deformation result is used; According to the preset threshold , , The degree of deformation is divided into four levels, let the deformation level be 1. Then we have: ; in, =0 indicates no significant deformation. =1 indicates slight deformation. =2 indicates moderate deformation. =3 indicates significant deformation. Normalizing the deformation levels yields the deformation weights. : ; Set category severity weights for: ; set up The deformation enhancement coefficient, The non-deformation suppression coefficient, Indicates whether deformation correction is enabled. When deformation correction is enabled... =1, deformation correction is not enabled. =0, then the prediction probability correction formula is: ; in, This is an intermediate probability during the correction process; Normalize the intermediate probabilities to obtain the final probabilities for detecting four types of changes. : ; in, Indicates the first step in the correction process The pixel belongs to the first Predicted probability of class changes.

[0018] Furthermore, the final change detection result is expressed as follows: ; in, Indicates the first The final change category of each pixel.

[0019] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement any of the above-described remote sensing image change detection methods with fused D-InSAR deformation supervision.

[0020] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: (1) This invention introduces D-InSAR deformation information into the remote sensing image change detection process to help determine whether the change has actually occurred, so that the change judgment not only depends on the surface difference of the optical image, but also combines the real deformation information of the ground surface.

[0021] (2) The present invention adopts a joint learning method of change detection and deformation estimation, so that the joint learning network learns optical change features and deformation features at the same time during the training process, thereby improving the joint learning network's ability to recognize real changes.

[0022] (3) In the reasoning stage, the present invention uses deformation information to correct the change detection results, which can reduce false changes caused by changes in lighting, shadows, seasonal differences, clouds and fog and imaging conditions, and improve the accuracy and reliability of building changes and infrastructure damage identification in disaster scenarios.

[0023] (4) The present invention can flexibly use D-InSAR deformation information according to the availability of SAR data. Deformation supervision can be added in the training stage, and deformation correction can be added in the inference stage. Finally, the changed area is divided into four categories: no obvious change, slight change, moderate change and significant change, which makes it more applicable.

[0024] (5) The present invention outputs four types of results: no obvious change, slight change, moderate change and significant change, which are more suitable for change identification, loss assessment and emergency analysis after disasters such as extreme rainfall, typhoon, flood, earthquake, etc., and can improve the accuracy and reliability of change detection in disaster scenarios such as extreme rainfall, typhoon, landslide, flood, ground subsidence, etc. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of a remote sensing image change detection method that integrates D-InSAR deformation supervision, provided by an embodiment of the present invention. Figure 2 This is a flowchart illustrating the training phase provided in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the reasoning stage provided in an embodiment of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0028] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0029] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0030] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0031] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0032] This invention provides a method for detecting changes in remote sensing images by fusing differential synthetic aperture radar interferometry (D-InSAR) deformation supervision, such as... Figure 1 As shown, the processing flow of this method may include the following steps: S1, acquire dual-temporal optical remote sensing images of the target area and corresponding synthetic aperture radar (SAR) images of the area; the dual-temporal optical remote sensing images are optical remote sensing images before and after the change. In this embodiment, the optical remote sensing image is used to provide information on the surface texture, color, and spatial structure of the target area before and after the change.

[0033] S2, if SAR images can be acquired, D-InSAR deformation information reflecting surface deformation can be generated; In this embodiment, when SAR images are available, D-InSAR deformation information is generated using SAR images from previous and subsequent time phases to reflect the surface subsidence, uplift, slippage, or other deformation of the target area before and after the change.

[0034] In this embodiment, the D-InSAR deformation information is vertical deformation information, which is used to reflect the vertical deformation of the target area before and after the change.

[0035] As an optional embodiment, to obtain the vertical deformation component, when the horizontal deformation of the target area surface is relatively small, the vertical component can be extracted based on the geometric relationship between the deformation along the radar line of sight and the incident angle. Let the deformation along the radar line of sight be... The radar incident angle is The vertical deformation is Then we have: ; Therefore, the vertical deformation can be expressed as: .

[0036] In another preferred embodiment, when simultaneously acquiring ascending and descending SAR images, the vertical and east-west deformation components can be decomposed by combining the line-of-sight deformation results of the ascending and descending orbits. Let the line-of-sight deformation of the ascending orbit be... The line of sight during the descent changes into deformation. The incident angles corresponding to the ascending and descending orbits are respectively , The azimuth angles are respectively , The vertical deformation is The east-west deformation is Then the following observation equation can be established: ; ; The vertical deformation component can be obtained by solving using least squares. and east-west deformation components For scenarios involving the main risk characteristics covered by this patent, vertical deformation is preferentially used as deformation monitoring information.

[0037] S3 preprocesses optical remote sensing images and D-InSAR deformation information, including image registration, cropping, resampling and normalization, to ensure that the two are consistent in spatial location, resolution and image range. S4. Construct a joint learning network consisting of a shared encoder, a change detection decoder, and a deformation estimation decoder. The shared encoder is used to extract the change features of the optical remote sensing images before and after the change, the change detection decoder is used to output four types of change detection probability maps, and the deformation estimation decoder is used to output the predicted deformation results. S5: Input the dual-temporal optical remote sensing images into the shared encoder to extract change features; In this embodiment, before inputting the dual-temporal optical remote sensing images into the shared encoder, the optical remote sensing images before and after the change are first input into two weighted optical feature encoders to obtain the pre-temporal optical features and the post-temporal optical features, respectively. Then, the pre-temporal and post-temporal optical features are cascaded to obtain dual-temporal cascaded features. These dual-temporal cascaded features are then input into the shared encoder for further feature extraction to obtain the change features. The change features include the spatial features, semantic features, and temporal difference features of the optical remote sensing images before and after the change.

[0038] For simplicity, let the optical remote sensing image before the change be... The changed optical remote sensing image is The shared encoder is E, and the extracted variation features are: ; in, This represents the change features extracted from optical remote sensing images before and after the change.

[0039] S6. Input the change features into the change detection decoder to obtain four types of change detection probability maps: no obvious change, slight change, moderate change, and significant change. In this embodiment, the change detection decoder is used to output the probability that each pixel belongs to no significant change, slight change, moderate change, and significant change, respectively. Let the change detection decoder be The resulting probability maps for detecting four types of changes are as follows: ; Wherein, Softmax represents the normalization exponential function, which is used to convert the response values ​​of each category output by the change detection decoder into a probability distribution, so that the sum of the predicted probabilities of each pixel in the four change categories is 1; Indicates the first The pixel belongs to the first Predicted probability of class change =0,1,2,3 This represents the change features extracted from optical remote sensing images before and after the change.

[0040] S7, input the change features into the deformation estimation decoder to obtain the predicted deformation result; In this embodiment, the deformation estimation decoder is used to estimate the degree of deformation at each pixel position in the target area, that is, to output the predicted deformation result, which is used to estimate the degree of deformation at each pixel position in the target area. Let the deformation estimation decoder be The predicted deformation results are as follows: ; in, Indicates the first The predicted deformation value of each pixel. This represents the change features extracted from optical remote sensing images before and after the change.

[0041] S8. During the training phase, based on the acquisition status of D-InSAR deformation information, it is selected whether to add deformation supervision: when D-InSAR deformation information is available, the joint learning network is trained using change ground truth labels and D-InSAR deformation information; when D-InSAR deformation information is unavailable, the change detection branch in the joint learning network is trained using change ground truth labels. The change detection branch includes a shared encoder for extracting change features from dual-temporal optical remote sensing images, a change detection decoder, and four types of change probability output layers.

[0042] like Figure 2As shown, during the training phase, when D-InSAR deformation information is available, the joint learning network is trained using the change ground truth label and D-InSAR deformation information, and optimized using a joint loss function including change detection loss and deformation estimation loss; whereby the change detection loss is used to constrain the four types of change detection results, and the deformation estimation loss is used to constrain the predicted deformation results. set up Indicates the first The pixel in the first truth labels of changes on the class Indicates the first The pixel belongs to the first Predicted probability of class change Indicates the number of pixels. To prevent abnormal logarithmic calculations or extremely small constants with a denominator of zero, the four-class cross-entropy loss is used. for: ; loss Used to measure the overlap between predicted results and true labels, and since changed areas are usually fewer than unchanged areas in remote sensing change detection, introducing Dice loss can enhance the model's focus on changed areas. Represented as: ; Therefore, change detection loss Represented as: ; in, Indicates loss Weighting coefficients; Deformation estimation loss is used to constrain the difference between the predicted deformation result and the D-InSAR deformation information, assuming Indicates the first D-InSAR deformation value of each pixel Indicates the first The predicted deformation value of each pixel. Indicates the first Does each pixel possess valid D-InSAR deformation information? If valid... =1, invalid =0, then the deformation estimation loss for: ; Therefore, the total loss function is expressed as: ; in, Indicates loss The weight coefficient s represents whether D-InSAR deformation supervision is added during the training phase; when D-InSAR deformation information is available, s=1; when D-InSAR deformation information is not available, s=0. In other words, when D-InSAR deformation information is available, the total loss function, i.e., the joint loss function, is: ; When D-InSAR deformation information is unavailable, the change detection branch is trained using the ground truth labels of the changes; that is, when D-InSAR deformation information is unavailable, the total loss is: .

[0043] S9 employs an adaptive deformation correction strategy during the inference phase: when D-InSAR deformation information is available, it is used to correct the probability maps of the four types of changes; when D-InSAR deformation information is unavailable, the predicted deformation results or the output results of the change detection decoder are used to generate the change detection map. like Figure 3 As shown, during the inference phase, when D-InSAR deformation information is available, the probability map for detecting four types of changes is corrected using the D-InSAR deformation information; when D-InSAR deformation information is unavailable, but the predicted deformation result is available, the predicted deformation result is used for correction; when neither is available, a change detection map is generated based on the output result of the change detection decoder; the correction includes adjusting the probability of detecting four types of changes according to the degree of deformation. When the degree of deformation is high, the probability of moderate or significant changes is increased; when the degree of deformation is low, the probability of moderate or significant changes is decreased.

[0044] Let the first The degree of deformation of each pixel is Then we have: ; in, =1 indicates that D-InSAR vertical deformation information is used. =0 indicates that the predicted deformation result is used; According to the preset threshold , , The degree of deformation is divided into four levels, let the deformation level be 1. Then we have: ; in, =0 indicates no significant deformation. =1 indicates slight deformation. =2 indicates moderate deformation. =3 indicates significant deformation. Normalizing the deformation levels yields the deformation weights. : ; in, The value range of is [0,1]. The more obvious the deformation, the more pronounced the deformation. The larger; To ensure that the results of all four types of changes can be corrected according to the degree of deformation, a category severity weight is set. for: ; set up The deformation enhancement coefficient, The non-deformation suppression coefficient, Indicates whether deformation correction is enabled. When deformation correction is enabled... =1, deformation correction is not enabled. =0, then the prediction probability correction formula is: ; in, This represents the intermediate probability during the correction process. The formula indicates that when deformation is significant, Wi is larger, and the probability of the change category is enhanced; when deformation is insignificant, Wi is smaller, and the probability of the change category is suppressed. Since qc increases with the level of change, moderate and significant changes are more significantly affected by deformation correction. Normalize the intermediate probabilities to obtain the final probabilities for detecting four types of changes. : ; in, Indicates the first step in the correction process The pixel belongs to the first Predicted probability of class changes.

[0045] S10, based on the corrected four-type change detection probability map or the output of the change detection decoder, the final change detection result is obtained, expressed as: ; in, Indicates the first The final change category of each pixel.

[0046] In this embodiment, the final change detection results are used for identifying building changes, infrastructure damage, and assessing disaster losses after disasters such as extreme rainfall, typhoons, floods, earthquakes, landslides, and fires.

[0047] In summary, the remote sensing image change detection method fused with D-InSAR deformation supervision provided in this embodiment of the invention has at least the following beneficial effects: (1) This invention introduces D-InSAR deformation information into the remote sensing image change detection process to help determine whether the change has actually occurred, so that the change judgment not only depends on the surface difference of the optical image, but also combines the real deformation information of the ground surface.

[0048] (2) The present invention adopts a joint learning method of change detection and deformation estimation, so that the joint learning network learns optical change features and deformation features at the same time during the training process, thereby improving the joint learning network's ability to recognize real changes.

[0049] (3) In the reasoning stage, the present invention uses deformation information to correct the change detection results, which can reduce false changes caused by changes in lighting, shadows, seasonal differences, clouds and fog and imaging conditions, and improve the accuracy and reliability of building changes and infrastructure damage identification in disaster scenarios.

[0050] (4) The present invention can flexibly use D-InSAR deformation information according to the availability of SAR data. Deformation supervision can be added in the training stage, and deformation correction can be added in the inference stage. Finally, the changed area is divided into four categories: no obvious change, slight change, moderate change and significant change, which makes it more applicable.

[0051] (5) The present invention outputs four types of results: no obvious change, slight change, moderate change and significant change, which are more suitable for change identification, loss assessment and emergency analysis after disasters such as extreme rainfall, typhoon, flood, earthquake, etc., and can improve the accuracy and reliability of change detection in disaster scenarios such as extreme rainfall, typhoon, landslide, flood, ground subsidence, etc.

[0052] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0053] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0054] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0055] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0056] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0058] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0059] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0060] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0061] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0062] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting changes in remote sensing images integrating D-InSAR deformation supervision, characterized in that, The method includes: S1, acquire dual-temporal optical remote sensing images of the target area and corresponding SAR images of the area; the dual-temporal optical remote sensing images are optical remote sensing images before and after the change. S2, if SAR images can be acquired, D-InSAR deformation information reflecting surface deformation can be generated; S3 preprocesses optical remote sensing images and D-InSAR deformation information to ensure that the two are consistent in spatial location, resolution and image range. S4, construct a joint learning network consisting of a shared encoder, a change detection decoder, and a deformation estimation decoder; S5: Input the dual-temporal optical remote sensing images into the shared encoder to extract change features; S6. Input the change features into the change detection decoder to obtain four types of change detection probability maps: no obvious change, slight change, moderate change, and significant change. S7, input the change features into the deformation estimation decoder to obtain the predicted deformation result; S8. During the training phase, based on the acquisition status of D-InSAR deformation information, select whether to add deformation supervision: when D-InSAR deformation information is available, use the change ground truth label and D-InSAR deformation information to train the joint learning network; when D-InSAR deformation information is not available, use the change ground truth label to train the change detection branch. S9 employs an adaptive deformation correction strategy during the inference phase: when D-InSAR deformation information is available, it is used to correct the probability maps of the four types of changes; when D-InSAR deformation information is unavailable, the predicted deformation results or the output results of the change detection decoder are used to generate the change detection map. S10: Based on the corrected four-type change detection probability map or the output of the change detection decoder, the final change detection result is obtained.

2. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 1, characterized in that, The optical remote sensing images are used to provide information on the surface texture, color, and spatial structure of the target area before and after changes. The preprocessing includes image registration, cropping, resampling, and normalization.

3. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 1, characterized in that, The D-InSAR deformation information is vertical deformation information, represented as follows: ； in, Represents vertical deformation. This indicates the deformation of the radar line-of-sight direction. Indicates the radar incident angle.

4. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 1, characterized in that, The D-InSAR deformation information is vertical deformation information; The following observation equation is established: ； ； Solving for the vertical deformation component using least squares. and east-west deformation components , To cause deformation of the line of sight during ascent, To reduce the deformation along the line of sight, , These are the incident angles for the ascending and descending orbits, respectively. , These are the azimuth angles corresponding to the ascending and descending orbits, respectively.

5. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 1, characterized in that, The shared encoder is used to extract the change features of optical remote sensing images before and after the change; Let the optical remote sensing image before the change be... The changed optical remote sensing image is The shared encoder is E, and the extracted variation features are: ； in, This represents the change features extracted from optical remote sensing images before and after the change.

6. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 1, characterized in that, The change detection decoder is used to output the probability that each pixel belongs to no significant change, slight change, moderate change, and significant change, respectively. Let the change detection decoder be The resulting probability maps for detecting four types of changes are as follows: ； in, Indicates the first The pixel belongs to the first Predicted probability of class change =0,1,2,3 This represents the change features extracted from the optical remote sensing images before and after the change, and Softmax represents the normalized exponential function.

7. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 1, characterized in that, The deformation estimation decoder is used to estimate the degree of deformation at each pixel position in the target area, that is, to output the predicted deformation result; Let the deformation estimation decoder be The predicted deformation results are as follows: ； in, Indicates the first The predicted deformation value of each pixel. This represents the change features extracted from optical remote sensing images before and after the change.

8. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 1, characterized in that, During the training phase, when D-InSAR deformation information is available, a joint learning network is trained using the change ground truth label and D-InSAR deformation information, and optimized using a joint loss function that includes change detection loss and deformation estimation loss; whereby the change detection loss is used to constrain the four types of change detection results, and the deformation estimation loss is used to constrain the predicted deformation results. set up Indicates the first The pixel in the first truth labels of changes on the class Indicates the first The pixel belongs to the first Predicted probability of class change Indicates the number of pixels. To prevent abnormal logarithmic calculations or extremely small constants with a denominator of zero, the four-class cross-entropy loss is used. for: ； loss The degree of overlap between the predicted result and the true label is used to measure the degree of overlap, and is expressed as: ； Change detection loss Represented as: ； in, Indicates loss Weighting coefficients; set up Indicates the first D-InSAR deformation value of each pixel Indicates the first The predicted deformation value of each pixel. Indicates the first Does each pixel possess valid D-InSAR deformation information? If valid... =1, invalid =0, then the deformation estimation loss for: ； The joint loss function is expressed as: ； in, Indicates loss The weighting coefficients.

9. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 8, characterized in that, During the inference phase, when D-InSAR deformation information is available, the probability map for four types of change detection is corrected using the D-InSAR deformation information; when D-InSAR deformation information is unavailable, but the predicted deformation result is available, the predicted deformation result is used for correction; when neither is available, a change detection map is generated based on the output result of the change detection decoder. Let the first The degree of deformation of each pixel is Then we have: ； in, =1 indicates that D-InSAR vertical deformation information is used. =0 indicates that the predicted deformation result is used; According to the preset threshold , , The degree of deformation is divided into four levels, let the deformation level be 1. Then we have: ； in, =0 indicates no significant deformation. =1 indicates slight deformation. =2 indicates moderate deformation. =3 indicates significant deformation. Normalizing the deformation levels yields the deformation weights. : ； Set category severity weights for: ； set up The deformation enhancement coefficient, The non-deformation suppression coefficient, Indicates whether deformation correction is enabled. When deformation correction is enabled... =1, deformation correction is not enabled. =0, then the prediction probability correction formula is: ； in, This is an intermediate probability during the correction process; Normalize the intermediate probabilities to obtain the final four types of change detection probabilities. : ； in, Indicates the first step in the correction process The pixel belongs to the first Predicted probability of class changes.

10. The remote sensing image change detection method fused with D-InSAR deformation supervision according to claim 9, characterized in that, The final change detection results are expressed as follows: ； in, Indicates the first The final change category of each pixel.

Images