Multi-modal remote sensing image change detection method and system and readable storage medium

Abstract

The invention relates to a multi-modal remote sensing image change detection method and system and a readable storage medium, and the method comprises the steps: carrying out the normalization of a multi-modal remote sensing image, carrying out the superpixel segmentation through employing simple linear iteration clustering, and employing the pixels of superpixels as nodes to construct an original image structure; inputting the original graph structure into a graph attention network to extract graph structure features of superpixels; performing style regularization on the graph structure features, and performing style randomization on the graph structure features and adjacent superpixels to obtain new style features of the superpixels; inputting the new style features into a graph structure decoder to obtain a super-pixel reconstructed graph structure; training a graph attention network by using the content consistency of the reconstructed graph structure and the original graph structure to obtain a target graph attention network; and inputting an original image structure obtained by processing a to-be-detected multi-modal remote sensing image before and after change into the target image attention network to obtain image structure features before and after change, and obtaining a change binary image by using an Otsu method. The change detection precision is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing image change detection, and in particular relates to a multimodal remote sensing image change detection method, system and readable storage medium. Background Art

[0002] With the rapid development of remote sensing technology, long-term observation data can be obtained for remote sensing image change detection and the detection of the evolution and change of the earth. Compared with single-modal remote sensing image change detection, multimodal remote sensing image change detection can achieve all-weather and all-day observations. By integrating multi-dimensional features such as spectrum, texture, and time series, it provides a richer observation perspective for surface change analysis. This data advantage not only brings new research opportunities, but also poses severe challenges to traditional change detection methods. Unsupervised learning technology has been applied to remote sensing image target detection tasks. It can not only break the dependence on a large amount of labeled data and solve the problem of scarce labels, but also fully explore the potential correlation and complementarity of multimodal data, and improve the accuracy and robustness of change detection. Unsupervised multimodal remote sensing image change detection has practical application value in the following two aspects: 1) Disaster assessment: Multi-time series remote sensing change detection can effectively monitor the evolution of geological disasters. By integrating multimodal data of SAR and optical images, it can break through the interference of clouds and fog to achieve millimeter-level monitoring of landslide displacement; 2) Environmental monitoring: As global climate change intensifies, timely understanding of surface cover evolution is crucial to ecological and environmental protection; for example, by detecting changes in ecologically sensitive areas such as deforestation and wetland degradation, it can provide data support for carbon sink assessment and biodiversity protection.

[0003] Due to the different imaging mechanisms of multimodal images, the features of the images are different, and change detection cannot be performed by direct comparison; moreover, multimodal image samples are naturally limited, resulting in a scarce number of available labels, and label annotation is costly. Existing unsupervised change detection methods for multimodal remote sensing images mainly mine the correlation between cross-modal images through feature alignment and modality conversion to achieve detection of changed areas; however, there are still problems such as missed detection or false alarm of changed areas due to global scale imaging differences of multimodal sensors, and discontinuity of changed areas due to local scale imaging disturbances of multimodal remote sensing images.

[0004] Regarding the problem of imaging differences in multi-modal sensors, for example, existing image regression methods based on structural periodic consistency transform one image into the domain of another image, compare the images, and then directly extract changes. This method aligns the domain space of the images, but the quality of the regression image will affect the detection accuracy. Another example is that existing technologies also fuse heterogeneous images and generate pre- and post-event phase images with consistent spectral and spatial domains based on the fusion results, and then directly measure the change amplitude. This method can align the features of optical images and synthetic aperture radar (SAR) images through data fusion strategies, but due to the modal differences of heterogeneous images, data anomalies are likely to occur after image fusion. In addition, existing technologies also use an affinity matrix to guide the image to a common space and then compare the features in this space. This method does not need to consider the modal differences between heterogeneous images, but the detection performance depends on the feature extraction effect of the feature learning network or algorithm.

[0005] Regarding the problem of establishing the context relationship of multi-modal images, many existing methods use graph structures to mine the shared features between images and can also evaluate the degree of change between heterogeneous images through graph similarity measurement. For example, a method of constructing a graph structure for multi-modal images effectively represents complex data and proposes a structural relationship graph convolutional autoencoder to obtain representative features between multi-modal remote sensing images. Another example is a graph structure learning method based on change alignment to optimize the learning process of the graph structure. This method has successfully improved the performance of change detection in multi-modal remote sensing images. However, due to the heterogeneity still existing in the graph structures constructed by different modal images and the domain differences between graph structures, there is still a problem of discontinuous change regions caused by local-scale imaging perturbations.

[0006] Although the above existing methods have improved the performance of change detection in multi-modal remote sensing images, it is still difficult to effectively extract the modal differences in the unsupervised case and multi-modal graph structures. Summary of the Invention

[0007] Based on the above-mentioned drawbacks and deficiencies existing in the prior art, one of the objectives of the present invention is to at least solve one or more of the above problems existing in the prior art. In other words, one of the objectives of the present invention is to provide a multi-modal remote sensing image change detection method, system, and readable storage medium that meet one or more of the foregoing requirements, based on unsupervised learning and graph structure style transfer.

[0008] To achieve the above-mentioned invention objective, the present invention adopts the following technical solutions: A multi-modal remote sensing image change detection method includes the following steps: S1. Normalize the multi-modal remote sensing images in the multi-modal remote sensing image training set, perform superpixel segmentation on the normalized multi-modal remote sensing images using simple linear iterative clustering, and use the pixels of the superpixels as the nodes of the graph structure to construct the original graph structure of each superpixel; wherein, the multi-modal remote sensing image training set includes the multi-modal remote sensing images before and after the change. S2. Input the original graph structure of the superpixels into the graph attention network to extract features, and obtain the graph structure features of the superpixels. S3. Perform style regularization on the graph structure features of the superpixels and perform style randomization with neighboring superpixels to obtain the new style features of the superpixels. S4. Input the new style features of the superpixels into the graph structure decoder to perform graph structure reconstruction, and obtain the reconstructed graph structure of the superpixels; use the content consistency between the reconstructed graph structure of the superpixels and the original graph structure to train the graph attention network to obtain the converged graph attention network. S5. Input the original graph structures obtained by processing the multi-modal remote sensing images before and after the change to be measured through the above step S1 into the converged graph attention network, obtain the graph structure features before and after the change, and use the Otsu method to obtain the binary change map.

[0009] As a preferred solution, the step S3 specifically includes the following steps: Step S31. Perform style regularization on the graph structure features of the current superpixel, and exchange the style information of the graph structures before and after the change. Step S32. Select neighboring superpixels and perform style randomization with the style information of the current superpixel to obtain the new style features of the current superpixel. Among them, the process of style randomization is: ; ; Among them, and are the mean and variance of the graph structure features of the current superpixel respectively, and are the mean and variance of the graph structure features of the neighboring superpixels respectively, is a random number, and are the randomly generated mean and variance respectively, that is, the new style features of the current superpixel.

[0010] As a preferred solution, in the step S31, use adaptive instance normalization to perform style exchange on the graph structures before and after the change.

[0011] As a preferred solution, the constraint condition for constructing the content consistency between the reconstructed graph structure of the superpixels and the original graph structure by using the L2 norm loss function is: ; where N is the number of superpixels, and respectively represent the original graph structures of the i-th superpixel of the multi-modal remote sensing images before and after the change, and respectively represent the reconstructed graph structures of the i-th superpixel of the multi-modal remote sensing images before and after the change after style randomization.

[0012] As a preferred solution, in the step S4, an unsupervised training graph attention network is performed by using the style change but content consistency constraint condition.

[0013] As a preferred solution, the multi-modal remote sensing images include optical images and synthetic aperture radar (SAR) images.

[0014] As a preferred solution, the step S1 specifically includes the following steps: S11. Perform min-max normalization on the optical image to obtain a normalized optical image; Perform logarithmic transformation on the SAR image, and then perform min-max normalization on the transformed image to obtain a normalized SAR image; S12. Use simple linear iterative clustering to perform superpixel segmentation on the normalized optical image, and apply the segmentation result to the normalized SAR image. Then, splice the normalized optical image and the normalized SAR image in the channel dimension to generate a unified superpixel segmentation result; S13. Use each pixel within the superpixel as a graph structure node to construct a graph structure, which is the original graph structure of the superpixel.

[0015] As a preferred solution, in the step S5, the difference between the graph structure features before and after the change is obtained to get a feature difference value, and the Otsu method is used to obtain a binary change map.

[0016] The present invention also provides a multi-modal remote sensing image change detection system, which applies the multi-modal remote sensing image change detection method described in any one of the above solutions. The multi-modal remote sensing image change detection system includes: A normalization module, which is used to normalize the multi-modal remote sensing images in the multi-modal remote sensing image training set; and is also used to normalize the multi-modal remote sensing images before and after the change to be measured; A superpixel segmentation module, which is used to perform superpixel segmentation on the normalized multi-modal remote sensing images by using simple linear iterative clustering; A graph structure construction module, which is used to construct the original graph structure of each superpixel by taking the pixels of the superpixel as the graph structure nodes; A graph attention network module, which is used to input the original graph structure of the superpixel into the graph attention network to extract features and obtain the graph structure features of the superpixel; A graph style transfer module, which is used to perform style regularization on the graph structure features of the superpixel and perform style randomization with neighboring superpixels to obtain the new style features of the superpixel; A graph structure decoder module, which is used to input the new style features of the superpixel into the graph structure decoder to reconstruct the graph structure and obtain the reconstructed graph structure of the superpixel; A training module, which is used to train the graph attention network by using the content consistency between the reconstructed graph structure and the original graph structure of the superpixel to obtain the converged graph attention network; A detection module, which is used to input the original graph structure into the converged graph attention network to obtain the graph structure features before and after the change, and use the Otsu method to obtain the binary change map.

[0017] The present invention also provides a readable storage medium, in which instructions are stored. When the instructions are run on a computer, the computer is made to execute the multi-modal remote sensing image change detection method described in any one of the above solutions.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention selects the style information of neighboring superpixels and the current superpixel for style randomization, so that the graph attention network model can learn rich style information, making the model more robust to style information and improving the change detection accuracy; (2) The present invention uses graph structure style transfer to perform style exchange on the graph structures of multi-modal remote sensing images before and after the change, and uses the graph structure content consistency constraint to train the graph attention network, so that the features it can extract contain more content-irrelevant information, effectively improving the accuracy of multi-modal remote sensing image change detection; (3) The present invention uses neighboring superpixels to randomly generate styles, and randomly generates and exchanges style features in the local neighborhood, so that adjacent superpixels can absorb the style information of neighboring regions while maintaining their own features. This can not only effectively alleviate the style mutation problem in the boundary region, but also enrich the feature representation of the superpixel and improve the continuity of the feature space. Description of the Drawings

[0019] Figure 1 is the flowchart of the multi-modal remote sensing image change detection method in Embodiment 1 of the present invention; Figure 2 is the module architecture diagram of the multi-modal remote sensing image change detection system in Embodiment 1 of the present invention; Figure 3It is the change detection result diagram of Embodiment 1 of the present invention. Detailed implementation manners

[0020] To more clearly illustrate the embodiments of the present invention, the specific implementation manners of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings, and other implementation manners can also be obtained.

[0021] Embodiment 1: The multi-modal remote sensing image change detection framework based on unsupervised learning and graph structure style transfer in this embodiment consists of three components: a graph attention network, a graph style transfer framework, and a decoder for reconstructing the graph structure; the specific multi-modal remote sensing image change detection process includes: First, use simple linear iterative clustering to perform superpixel segmentation on the multi-modal remote sensing image, construct a graph structure with each pixel of the superpixel as a node, and input it into the graph attention network for feature extraction; Second, the graph style transfer framework performs style regularization on the extracted features above, randomly generates a new style with adjacent superpixels and replaces the original style of the features, so that the network can learn more styles, and thus the extracted features contain more style-independent knowledge; Third, input the features of the new style into the graph structure decoder to reconstruct the graph structure. The reconstructed graph structure should contain the same content as the graph structure input to the graph attention network, and use the graph structure content consistency to train the graph attention network; Finally, in the detection stage, use the features extracted by the trained and converged graph attention network to perform change detection.

[0022] As Figure 1 shown, the multi-modal remote sensing image change detection method based on unsupervised learning and graph structure style transfer in this embodiment includes the following steps: (1) Collect the multi-modal remote sensing image before change and the multi-modal remote sensing image after change, and construct a multi-modal remote sensing image training set; wherein, the multi-modal remote sensing image before change is simply referred to as the image X before change, and the multi-modal remote sensing image after change is simply referred to as the image Y after change; The multi-modal remote sensing image in this embodiment includes an optical image and a synthetic aperture radar (SAR) image.

[0023] (2) Normalize the multi-modal remote sensing images in the multi-modal remote sensing image training set, use simple linear iterative clustering to perform superpixel segmentation on the normalized multi-modal remote sensing images, and construct the original graph structure of each superpixel with the pixels of the superpixel as graph structure nodes; For optical images, the minimum-maximum normalization method is used for normalization to obtain normalized optical images. For SAR images, first, the SAR images need to be logarithmically transformed to suppress speckle noise, and then the transformed images are subjected to minimum-maximum normalization to obtain normalized SAR images. Next, simple linear iterative clustering is used for superpixel segmentation and a co-segmentation strategy is adopted. The multimodal image pairs are stacked in the channel dimension to obtain a unified segmentation result. Specifically, simple linear iterative clustering is used to perform superpixel segmentation on the normalized optical images, and the segmentation results are applied to the normalized SAR images. Then, the normalized optical images and the normalized SAR images are concatenated in the channel dimension to generate a unified superpixel division result. Finally, based on the superpixels obtained by segmentation, each pixel within the superpixels is used as a node of the graph structure to construct a graph structure, obtaining the structural information of the multimodal remote sensing images as the original graph structure of the superpixels.

[0024] (3)Build a graph attention network GAT, and input the original graph structure of the superpixels into the graph attention network to extract features, obtaining the graph structure features of the superpixels, that is, in this embodiment, based on the above-mentioned pre-change image X and post-change image Y, the pre-change image features F X and the post-change image features F Y can be obtained, and then the process of graph style knowledge transfer in step (4) below is carried out; Among them, the graph attention network can efficiently extract the features of the graph structure. In addition, in order to improve the accuracy of establishing the connection relationship between the graph vertices and improve the ability of the graph neural network to model the image structure features, an attention mechanism GAM is introduced. The attention mechanism can refer to the existing technology and will not be elaborated here.

[0025] (4)Perform style regularization on the graph structure features of the superpixels. Then, select neighboring superpixels and the current superpixel for style randomization, and use the graph structure content consistency to train the graph attention network to improve the ability of the network model to extract style-irrelevant features. Among them, the process of style regularization can refer to the existing technology and will not be elaborated here; In this embodiment, after extracting the graph structure features, style regularization is performed on the features, and the style information of the graph structures before and after the change is exchanged, and unsupervised training is carried out using the constraint of style change but content consistency. Although change detection of structural information by constructing graph structures for multimodal remote sensing images can reduce modal differences, for images with change detection accuracy, the graph structures constructed from different modal remote sensing images still face modal differences, resulting in false alarms for features with obvious imaging differences in different modalities. Based on this, in this embodiment, graph structure style transfer is introduced to exchange the styles of graph structures before and after change, and a graph attention network is trained using content consistency constraints so that the features it can extract contain more content-irrelevant information; Specifically, adaptive instance normalization is used to exchange the styles of graph structures before and after change; Then, neighboring superpixels are selected and their style information is randomly mixed with the style information of the current superpixel. Designed in this way, the network model can learn rich style information and make the model more robust to style information.

[0026] Since the complexity of features in remote sensing images also poses challenges to change detection, the sudden change of spectral features at the boundaries of features, when superpixel segmentation is used to process complex scenes such as water-land intersections, often produces obvious dividing lines, resulting in significant differences in style features between adjacent superpixels, which may introduce false alarm problems in the boundary area and thus affect the accuracy of subsequent change detection. Based on this, in this embodiment, random generation of neighboring superpixels is introduced, and random generation and exchange of style features are performed in the local neighborhood, so that adjacent superpixels can absorb the style information of neighboring regions while maintaining their own features; this strategy can not only effectively alleviate the style mutation problem in the boundary area, but also enrich the feature representation of superpixels and improve the continuity of the feature space.

[0027] Taking the feature F of the pre-change image in this embodiment X as an example, the process of style randomization is described in detail as follows:

[0028] ;

[0029] ; where and are the mean and variance of the graph structure features of the current superpixel respectively, and are the mean and variance of the graph structure features of the neighboring superpixel respectively, is a random number, and are the randomly generated mean and variance respectively, that is, the new style features of the current superpixel; The feature content after style randomization should be consistent with that before transformation. Therefore, the graph structure decoder before transformation and the graph structure decoder after transformation are respectively used to reconstruct the graph structure. The reconstructed graph structure should be content-consistent with the graph structure of the input graph attention network. The constraint condition for constructing the content consistency between the reconstructed graph structure of the superpixels and the original graph structure using the L2 norm loss function is as follows: ; where N is the number of superpixels, and respectively represent the original graph structures of the i-th superpixel of the multi-modal remote sensing images before and after transformation, and respectively represent the reconstructed graph structures of the i-th superpixel of the multi-modal remote sensing images before and after transformation after style randomization; Through the above steps (1) to (4), the training of the graph attention network is realized. After the training converges, the converged graph attention network is obtained. Among them, in this embodiment, the graph attention network is trained unsupervised using the constraint condition of style change but content consistency.

[0030] (5) Based on the converged graph attention network, change detection is performed on the sample to be measured, and the Otsu method is used to obtain the binary change detection result; Specifically, the original graph structures before and after transformation obtained by processing the multi-modal remote sensing images before and after the change to be measured through the above step (2) are input into the converged graph attention network to obtain the graph structure features before and after the change. The graph structure features before and after the change are subtracted to obtain the feature difference value, and the Otsu method is used to obtain the change binary graph.

[0031] Based on the above multi-modal remote sensing image change detection method, as Figure 2 shown, the multi-modal remote sensing image change detection system provided in this embodiment includes the following functional modules: a normalization module, a superpixel segmentation module, a graph structure construction module, a graph attention network module, a graph style transfer module, a graph structure decoder module, a training module, and a detection module; The normalization module of this embodiment is used to normalize the multi-modal remote sensing images in the multi-modal remote sensing image training set, and is also used to normalize the multi-modal remote sensing images before and after the change to be measured; The superpixel segmentation module of this embodiment is used to perform superpixel segmentation on the normalized multi-modal remote sensing images using simple linear iterative clustering; The graph structure construction module of this embodiment is used to construct the original graph structure of each superpixel with the pixels of the superpixel as the graph structure nodes; The graph attention network module of this embodiment is used to input the original graph structure of superpixels into the graph attention network to extract features, and obtain the graph structure features of superpixels; The graph style transfer module of this embodiment is used to perform style regularization on the graph structure features of superpixels, and perform style randomization with adjacent superpixels to obtain the new style features of superpixels; The graph structure decoder module of this embodiment is used to input the new style features of superpixels into the graph structure decoder to reconstruct the graph structure and obtain the reconstructed graph structure of superpixels; The training module of this embodiment is used to train the graph attention network by using the content consistency between the reconstructed graph structure and the original graph structure of superpixels, and obtain the converged graph attention network; The detection module of this embodiment is used to input the original graph structure into the converged graph attention network to obtain the graph structure features before and after the change, and use the Otsu method to obtain the binary change map; The specific processing process of the above functional modules can refer to the specific description in the above multi-modal remote sensing image change detection method, which will not be elaborated here.

[0032] The readable storage medium of this embodiment stores instructions. When the instructions run on a computer, the computer executes the above multi-modal remote sensing image change detection method to realize the intelligence of multi-modal remote sensing image change detection.

[0033] Applying the multi-modal remote sensing image change detection method of this embodiment to specific practical applications, the specific process is as follows: (1) Use the publicly available multi-modal change detection dataset Shuguang dataset. The Shuguang training dataset consists of SAR images and optical images; the size of the preprocessed SAR and optical images is 921×593 pixels; (2) Build a graph attention network and input the graph structure of multi-modal images into the network to extract features; Build a graph attention network to extract features. Use the existing commonly used three-layer graph convolutional layer to build a graph convolutional neural network. The network structure is GCN-GAT-GCN to extract the depth information in the graph structure and obtain the graph structure features; (3) Perform style randomization training on the extracted features for the network model, and use the network model for change detection, as Figure 3 shown, the white part is the changed part, and the black part is the unchanged part; the green area is the part where the change occurred but was not detected, which is the missed detection part; the red area detects the unchanged part as changed, which is the false alarm part.

[0034] The detection method of the present invention was compared and tested with the existing commonly used detection methods GIR_MRF, SCACS, IRG_McS, and SRGCAE. As shown in Table 1, the detection method of the present invention is superior to the existing detection methods in terms of the mean intersection over union (mIoU) index. In particular, compared with the existing SCACS algorithm, it is 17.3% higher in the mIoU index; compared with SRGCAE, which is also a deep learning-based algorithm, it is 4.2% higher in the mean intersection over union index. In addition, in terms of the foreground intersection over union (IoU_1) and the background intersection over union (IoU_2), the present invention shows the best performance. It is 32.7% better than the SCACS algorithm in IoU_1 and 8.3% better than the second-best SRGCAE algorithm. Although the present invention is slightly lower than the GIR_MRF algorithm and the IRG_McS algorithm in terms of the F1 score and the Kappa coefficient, in terms of the IoU_1 index, the present invention is 18.2% higher than the GIR_MRF algorithm and 14.6% higher than the IRG_McS algorithm.

[0035] Table 1 Comparative analysis indicators of different detection methods 。

[0036] The above description only details the preferred embodiments and principles of the present invention. For those of ordinary skill in the art, based on the ideas provided by the present invention, there will be changes in the specific implementation manners, and these changes should also be regarded as the protection scope of the present invention.

Claims

1. A multi-modal remote sensing image change detection method, characterized in that It includes the following steps: S1. Normalize the multi-modal remote sensing images in the multi-modal remote sensing image training set, perform superpixel segmentation on the normalized multi-modal remote sensing images using simple linear iterative clustering, and construct the original graph structure of each superpixel with the pixels of the superpixels as the graph structure nodes; among them, the multi-modal remote sensing image training set includes the multi-modal remote sensing images before and after the change; S2. Input the original graph structure of the superpixels into the graph attention network to extract features and obtain the graph structure features of the superpixels; S3. Perform style regularization on the graph structure features of the superpixels and perform style randomization with neighboring superpixels to obtain the new style features of the superpixels; S4. Input the new style features of the superpixels into the graph structure decoder for graph structure reconstruction to obtain the reconstructed graph structure of the superpixels; use the content consistency between the reconstructed graph structure of the superpixels and the original graph structure to train the graph attention network to obtain the converged graph attention network; S5. Input the original graph structures obtained by processing the multi-modal remote sensing images before and after the change to be measured through the above step S1 into the converged graph attention network to obtain the graph structure features before and after the change, and use the Otsu method to obtain the binary change map.

2. The multimodal remote sensing image change detection method according to claim 1, wherein The specific steps of step S3 include the following steps: Step S31. Perform style regularization on the graph structure features of the current superpixel and exchange the style information of the graph structures before and after the change; Step S32. Select neighboring superpixels and perform style randomization with the style information of the current superpixel to obtain the new style features of the current superpixel; Among them, the process of style randomization is: ； ； Among them, and are the mean and variance of the graph structure features of the current superpixel respectively, and are the mean and variance of the graph structure features of the neighboring superpixels respectively, is a random number, and are the randomly generated mean and variance respectively, that is, the new style features of the current superpixel.

3. The multi-modal remote sensing image change detection method according to claim 2, wherein In step S31, use adaptive instance normalization to perform style exchange on the graph structures before and after the change.

4. The multi-modal remote sensing image change detection method according to any one of claims 1-3, characterized in that, The constraint condition for constructing the content consistency between the reconstructed graph structure of the superpixels and the original graph structure using the L2 norm loss function is: ； where N is the number of superpixels, and respectively represent the original graph structures of the i-th superpixel of the multi-modal remote sensing images before and after the change, and respectively represent the reconstructed graph structures of the i-th superpixel of the multi-modal remote sensing images before and after the change after style randomization.

5. The multi-modal remote sensing image change detection method according to claim 4, characterized in that, In step S4, use the constraint condition of style change but content consistency to perform unsupervised training on the graph attention network.

6. The multi-modal remote sensing image change detection method according to any one of claims 1-3, characterized in that The multi-modal remote sensing images include optical images and synthetic aperture radar (SAR) images.

7. The multi-modal remote sensing image change detection method according to claim 6, wherein, The specific steps of step S1 include the following steps: S11. Perform min-max normalization on the optical images to obtain normalized optical images; Perform logarithmic transformation on the SAR images, and then perform min-max normalization on the transformed images to obtain normalized SAR images; S12. Perform superpixel segmentation on the normalized optical images using simple linear iterative clustering, apply the segmentation results to the normalized SAR images, and then splice the normalized optical images and the normalized SAR images in the channel dimension to generate a unified superpixel segmentation result; S13. Use each pixel within the superpixels as graph structure nodes to construct a graph structure as the original graph structure of the superpixels.

8. The multi-modal remote sensing image change detection method according to any one of claims 1-3, characterized in that In step S5, subtract the graph structure features before and after the change to obtain the feature difference, and use the Otsu method to obtain the binary change map.

9. A multi-modal remote sensing image change detection system, which applies the multi-modal remote sensing image change detection method according to any one of claims 1-8, characterized in that The multi-modal remote sensing image change detection system includes: A normalization module, which is used to normalize the multi-modal remote sensing images in the multi-modal remote sensing image training set; and is also used to normalize the multi-modal remote sensing images before and after the change to be measured; A superpixel segmentation module for performing superpixel segmentation on the normalized multimodal remote sensing image using simple linear iterative clustering; A graph structure construction module for constructing an original graph structure of each superpixel with the pixels of the superpixel as graph structure nodes; A graph attention network module for inputting the original graph structure of the superpixel into the graph attention network to extract features and obtaining the graph structure features of the superpixel; A graph style transfer module for performing style regularization on the graph structure features of the superpixel and performing style randomization with neighboring superpixels to obtain new style features of the superpixel; A graph structure decoder module for inputting the new style features of the superpixel into the graph structure decoder to reconstruct the graph structure and obtaining the reconstructed graph structure of the superpixel; A training module for training the graph attention network using the content consistency between the reconstructed graph structure and the original graph structure of the superpixel to obtain a converged graph attention network; A detection module for inputting the original graph structure into the converged graph attention network to obtain the graph structure features before and after the change and using Otsu's method to obtain the binary change map.

10. A readable storage medium stores instructions therein, characterized in that, When the instruction runs on a computer, it causes the computer to execute the multimodal remote sensing image change detection method according to any one of claims 1-8.

Images