Super-neighborhood sensing remote sensing image change detection method and device, equipment and medium

Abstract

The invention discloses a super-neighborhood perception remote sensing image change detection method, device and equipment and a medium, and belongs to the technical field of remote sensing image processing, and the method comprises the steps: obtaining a high-resolution remote sensing image, and extracting the multi-scale features of the high-resolution remote sensing image; extracting neighborhood context features to model context information; embedding a super-neighborhood context sensing position, combining the super-neighborhood context sensing position into a neighborhood context feature, adjusting the position weight of a super-neighborhood context image, and synchronizing the relationship between the multi-scale feature and the neighborhood context feature; capturing long-distance semantic correlation; capturing a long-distance dependency relationship between a neighborhood context feature and a multi-scale feature; combining the dual-time adjacent domain context features and the multi-scale features, and establishing time correlation; and constructing a super-neighborhood context sensing Transform network, and detecting a changing region in the dual-tense image. According to the method, the limitation of a remote sensing image sample cutting window is overcome, and the identification capability of large-scale change is enhanced.

Description

Technical Field

[0001] The invention relates to a super-neighborhood perception remote sensing image change detection method, device, equipment and medium, belonging to the technical field of remote sensing image processing. Background Art

[0002] Remote sensing image change detection identifies and analyzes changes in the Earth's surface by comparing remote sensing images of the same area at different times. It is crucial for urban development, environmental monitoring, disaster response, and resource management, and can provide timely and accurate information on surface changes. High-resolution remote sensing images provide rich ground object information and can accurately capture surface changes, thereby expanding their application range. However, this improved resolution has brought new challenges to change detection technology. In high-resolution remote sensing images, significant scale differences in land use and land cover changes are obvious. Traditional methods face challenges such as manual feature extraction and threshold adjustment, which limit their application. Deep learning, especially CNN, has greatly improved the performance of remote sensing change detection with its robust feature learning and adaptive model structure. Much research has been done to enhance the multi-scale feature representation of the network and improve the accuracy and applicability of deep learning change detection methods. However, the limited receptive field of CNN restricts them to only process small local areas, hindering the capture of broader spatial relationships. Therefore, ongoing research aims to overcome these limitations and ensure that CNN-based methods continue to develop for more accurate change detection.

[0003] Although various methods have been proposed to expand the receptive field in context modeling, they are still limited by the scope of sample cropping and are difficult to effectively capture ultra-long-range context information. Advanced models that integrate long-range context information from super-neighborhood samples are essential for enhancing context modeling, especially for remote sensing impact change detection. In remote sensing change detection, large-scale land use and land cover changes are issues of concern. In high-resolution images, the key information for identifying large-scale changes is usually beyond the scope of a single sample.

[0004] Traditional cropped sample-based models cannot overcome the sample range limitations in context perception. Therefore, it is urgent to overcome the sample cropping window limitations, effectively integrate the information of multiple adjacent samples, enhance the model's ability to recognize large-scale changes, and develop a change detection network for identifying large-scale changes. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a super-neighborhood perception remote sensing image change detection method, device, equipment and medium.

[0006] The technical solution adopted by the present invention to solve the technical problem is: In a first aspect, an embodiment of the present invention provides a super-neighborhood-aware remote sensing image change detection method, comprising the following steps: Step S1, obtaining high-resolution remote sensing images and extracting their multi-scale features; Step S2, using a convolutional neural network encoder to process the super-neighborhood context image and extract neighborhood context features to model context information; Step S3, embedding the super-neighborhood context-aware position, merging the relative position data into the neighborhood context features, adaptively adjusting the position weight of the super-neighborhood context image through HCTNet, and synchronizing the relationship between the multi-scale features and the neighborhood context features; Step S4, introduce a multi-head attention mechanism based on Vision Transformer to capture long-distance semantic correlation; Step S5, capturing the long-distance dependency between neighborhood context features and multi-scale features by stacking the super-neighborhood context attention module of the Transformer model; Step S6, fusing the super-neighborhood context attention module, using three super-neighborhood context attention modules to iteratively merge the bi-temporal neighboring domain context features and multi-scale features to establish temporal correlation; Step S7, constructing a super-neighborhood context-aware Transformer network based on steps S2 to S6, and using the super-neighborhood context-aware Transformer network to detect changed areas in the bi-temporal image.

[0007] As a possible implementation of this embodiment, the step S1, obtaining a high-resolution remote sensing image and extracting its multi-scale features, includes: Obtain several high-resolution remote sensing images from satellites or other remote sensing platforms. The remote sensing images obtained can be different bands or time series data of the same period; The acquired remote sensing images are rectified, radiometrically calibrated and geometrically corrected to eliminate atmospheric effects and camera deviations, so that they can be accurately positioned in the same geographic space; The ResNet34 network is used to extract multi-scale features from the processed remote sensing images to capture detailed local features that are crucial for accurate change detection for subsequent change detection.

[0008] As a possible implementation of this embodiment, step S2 uses a convolutional neural network encoder to process the super-neighborhood context image, extract neighborhood context features, and model context information, including: Downsample the super-neighborhood context image to a quarter of its original resolution while retaining key context information; A dual CNN encoder is used to perceive a wider context and extract neighborhood context features.

[0009] As a possible implementation of this embodiment, step S3, embedding the super-neighborhood context-aware position, merging the relative position data into the neighborhood context feature, adaptively adjusting the position weight of the super-neighborhood context image through HCTNet, and synchronizing the relationship between the multi-scale feature and the neighborhood context feature, includes: Extracting and fusing bi-temporal image features from remote sensing images using Conv-BN-ReLU blocks and , obtain the bitemporal difference feature ∈ ; Expand the bi-temporal difference features in the spatial dimension to generate a two-dimensional feature map ∈ ; The two-dimensional feature map Input into MLP, assign different influence levels to the two-dimensional features according to their distribution, and expand their spatial dimensions to obtain the bi-temporal difference features Spatially distributed attention maps of the same size ; Bitemporal difference feature With spatially distributed attention map Multiply them to obtain the change scale factor σ; Based on the change scale factor σ, HCTNet adaptively adjusts the position weight of the super-neighborhood context image and the relationship between multi-scale features and neighborhood context features.

[0010] As a possible implementation of this embodiment, in step S4, a multi-head attention mechanism is introduced based on Vision Transformer to capture long-distance semantic correlation; Capturing domain context features and multi-scale features , and optionally domain context features Projection to multi-scale features ; Extracting domain context features Contextual keywords and its value , obtain long-distance semantic correlation and use it to guide the Vision Transformer model to learn extensive contextual information.

[0011] As a possible implementation of this embodiment, step S5, capturing the long-distance dependency between the neighborhood context features and the multi-scale features by stacking the super-neighborhood context attention module of the Transformer model, includes: Swapping bi-temporal features after two iterations of super-neighborhood contextual attention modules and ; Perform a third iteration of the super-neighborhood contextual attention module to concatenate the bi-temporal features using the initial results and swapped features from each temporal branch and To exchange information, we can obtain enhanced dual-time features and , the enhanced dual-time feature and Contains rich contextual information and temporal relevance.

[0012] As a possible implementation of this embodiment, step S6, fusing the super-neighborhood context attention module, using three super-neighborhood context attention modules to iteratively merge the bi-temporal neighboring domain context features and the multi-scale features to establish time correlation, includes: The super-neighborhood context attention module is integrated, and three super-neighborhood context attention modules are used to iteratively merge the bi-temporal neighboring domain context features. and multi-scale features ; Based on the merged dual temporal adjacent domain context features and multi-scale features , establish time correlation.

[0013] As a possible implementation of this embodiment, the super-neighborhood context-aware Transformer network includes: Super-Neighborhood Context Branch (HCB) structure, which is used to aggregate extensive context information, overcome the limitation of cropping window and expand the receptive field of the network; Super Neighborhood Context-Aware Position Embedding (HCPE) is used to adaptively integrate neighborhood information, adaptively determine the relative optimal relative position weight encoding, determine the change scale factor by analyzing the dual-phase image, and capture the change scale of the core sample; The Hyper-Neighborhood Context Transformer (HCT) structure is used to fuse multi-scale features and neighborhood context features. It is built on the multi-head attention mechanism introduced by the visual transformer to capture long-range semantic correlations.

[0014] In a second aspect, an embodiment of the present invention provides a remote sensing image change detection device based on super-neighborhood perception, comprising: Data acquisition module, used to acquire high-resolution remote sensing images and extract their multi-scale features; A feature extraction module for processing the super-neighborhood context image using a convolutional neural network encoder to extract neighborhood context features to model context information; A data merging module for embedding super-neighborhood context-aware positions, merging relative position data into neighborhood context features, adaptively adjusting the position weights of super-neighborhood context images through HCTNet, and synchronizing the relationship between multi-scale features and neighborhood context features; The mechanism introduction module is used to introduce the multi-head attention mechanism based on Vision Transformer to capture long-distance semantic correlation; HCAM (Hyper-neighborhood Context Attention Modules) stacking module, which is used to capture the long-distance dependencies between neighborhood context features and multi-scale features by stacking the hyper-neighborhood context attention modules of the Transformer model; The feature merging module is used to fuse the super-neighborhood contextual attention module. Three super-neighborhood contextual attention modules are used to iteratively merge the bi-temporal neighboring domain contextual features and multi-scale features to establish temporal correlation. The image detection module is used to build a super-neighborhood context-aware Transformer network and use the super-neighborhood context-aware Transformer network to detect the changed areas in the bi-temporal image.

[0015] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of any of the above-mentioned super-neighborhood perception remote sensing image change detection methods.

[0016] In a fourth aspect, an embodiment of the present invention provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned super-neighborhood-aware remote sensing image change detection methods are executed.

[0017] The beneficial effects of the technical solution of the embodiment of the present invention are as follows: A super-neighborhood-aware remote sensing image change detection method according to the technical solution of an embodiment of the present invention comprises the following steps: step S1, obtaining a high-resolution remote sensing image and extracting its multi-scale features; step S2, processing a super-neighborhood context image using a convolutional neural network encoder, extracting neighborhood context features, and modeling context information; step S3, embedding a super-neighborhood context-aware position, merging relative position data into neighborhood context features, and adaptively adjusting the position weight of the super-neighborhood context image through HCTNet, as well as the relationship synchronization between the multi-scale features and the neighborhood context features; step S4, based on Vision Transformer introduces a multi-head attention mechanism to capture long-distance semantic correlation; step S5, by stacking the super-neighborhood context attention module of the Transformer model to capture the long-distance dependency between neighborhood context features and multi-scale features; step S6, fuse the super-neighborhood context attention module, use three super-neighborhood context attention modules to iteratively merge the bi-temporal neighborhood context features and multi-scale features, and establish temporal correlation; step S7, based on steps S2 to S6, construct a super-neighborhood context-aware Transformer network, and use the super-neighborhood context-aware Transformer network to detect the changed areas in the bi-temporal image. The present invention uses the constructed super-neighborhood context-aware Transformer network to perform high-resolution remote sensing image change detection, which can not only overcome the limitations of the remote sensing image sample cropping window, effectively integrate the information of multiple adjacent samples, but also enhance the recognition ability of large-scale changes in remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of a super-neighborhood-aware remote sensing image change detection method according to an exemplary embodiment; Figure 2 It is a schematic structural diagram of a super-neighborhood perception remote sensing image change detection device according to an exemplary embodiment. DETAILED DESCRIPTION

[0019] In order to more clearly illustrate the technical features of the solution of the present invention, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0020] like Figure 1 As shown, a super-neighborhood-aware remote sensing image change detection method provided by an embodiment of the present invention includes the following steps: Step S1, obtaining high-resolution remote sensing images and extracting their multi-scale features; Step S2, using a convolutional neural network encoder to process the super-neighborhood context image and extract neighborhood context features to model context information to enhance the network's ability to perceive spatial relationships in the surrounding environment; Step S3, embedding the super-neighborhood context-aware position, merging the relative position data into the neighborhood context features, adaptively adjusting the position weight of the super-neighborhood context image through HCTNet, and synchronizing the relationship between multi-scale features and neighborhood context features, thereby minimizing the interference of irrelevant context information; Step S4, introduce a multi-head attention mechanism based on Vision Transformer to capture long-distance semantic correlation; Step S5, capturing the long-distance dependencies between neighborhood context features and multi-scale features by stacking the Hyper-Neighborhood Context Attention Module (HCAM) of the Transformer model; Step S6, integrating the super-neighborhood context attention module, using three super-neighborhood context attention modules to iteratively merge the bi-temporal neighboring domain context features and multi-scale features, establish temporal correlation, and reduce the gap between the bi-temporal image domains; Step S7, constructing a super-neighborhood context-aware Transformer network based on steps S2 to S6, and using the super-neighborhood context-aware Transformer network to detect changed areas in the bi-temporal image.

[0021] As a possible implementation of this embodiment, the step S1, obtaining a high-resolution remote sensing image and extracting its multi-scale features, includes: Obtain several high-resolution remote sensing images from satellites or other remote sensing platforms. The remote sensing images obtained can be different bands or time series data of the same period; The acquired remote sensing images are rectified, radiometrically calibrated and geometrically corrected to eliminate atmospheric effects and camera deviations, so that they can be accurately positioned in the same geographic space; The ResNet34 network is used to extract multi-scale features from the processed remote sensing images to capture detailed local features that are crucial for accurate change detection, with a focus on capturing detailed local features that are crucial for accurate change detection for subsequent change detection.

[0022] The ResNet34 network consists of the following parts: an initial 7x7 convolutional layer for extracting features; four convolutional groups (conv2_x, conv3_x, conv4_x, conv5_x), each group contains multiple residual blocks, the first residual block of each convolutional group is usually a dotted residual block, which is used to adjust the shape of the feature matrix, a global average pooling layer and a fully connected layer for classification tasks.

[0023] As a possible implementation of this embodiment, step S2 uses a convolutional neural network encoder to process the super-neighborhood context image, extract neighborhood context features, and model context information, including: In order to efficiently extract generalized neighborhood context features , downsample the super-neighborhood context image to reduce its resolution to a quarter of its original size, ensuring computational efficiency and retaining key environmental information; A dual CNN encoder with a ResNet-like architecture is used to perceive a wider context and extract neighborhood context features to enhance its understanding of local semantic information.

[0024] The structure of the dual CNN encoder similar to the ResNet architecture is as follows: The encoder is divided into four layers to gradually capture and integrate comprehensive information about the context of distant neighbors 5. Each layer contains two convolutional blocks, each of which has a 3×3 convolutional layer (Conv), a batch normalization (BN) layer, and an activation function (ReLU). As the network progresses, the number of output channels increases to 64, 128, 256, and 512, while the output spatial resolution decreases to 1, 1 / 2, 1 / 4, and 1 / 8 of the input image size, respectively.

[0025] As a possible implementation of this embodiment, step S3, embedding the super-neighborhood context-aware position, merging the relative position data into the neighborhood context feature, adaptively adjusting the position weight of the super-neighborhood context image through HCTNet, and synchronizing the relationship between the multi-scale feature and the neighborhood context feature, includes: Extracting and fusing bi-temporal image features from remote sensing images using Conv-BN-ReLU blocks and , obtain the bitemporal difference feature ∈ ; Expand the bi-temporal difference features in the spatial dimension to generate a two-dimensional feature map ∈ ; The two-dimensional feature map Input into MLP, learn the impact of spatial layout changes, assign different influence levels to two-dimensional features according to the distribution of features, and expand its spatial dimension to obtain the bi-temporal difference features Spatially distributed attention maps of the same size ; Bitemporal difference feature With spatially distributed attention map Multiply them to obtain the change scale factor σ; Based on the change scale factor σ, HCTNet adaptively adjusts the position weight of the super-neighborhood context image and the relationship between multi-scale features and neighborhood context features.

[0026] By analyzing the bi-temporal images Determine the variation scaling factor σ to capture the proportional variation of the core samples.

[0027] As a possible implementation of this embodiment, in step S4, a multi-head attention mechanism is introduced based on Vision Transformer to capture long-distance semantic correlation; Capturing domain context features and multi-scale features , and optionally domain context features Projection to multi-scale features , local features are enhanced with comprehensive global context information; Extracting domain context features Contextual keywords and its value , obtain long-distance semantic correlation, and use long-distance semantic correlation to guide the Vision Transformer model to learn extensive contextual information, so that Gain a deeper and more comprehensive understanding of the context.

[0028] As a possible implementation of this embodiment, step S5, capturing the long-distance dependency between neighborhood context features and multi-scale features by stacking a hyper-neighborhood context attention module (HCAM) of a Transformer model, includes: The super-neighborhood contextual Transformer consists of two weight-sharing HCAMs and one fused HCAM, exchanging bi-temporal features after two super-neighborhood contextual attention module iterations. and , to ensure that the features in each time branch retain their bi-temporal characteristics; Perform a third iteration of the super-neighborhood contextual attention module to concatenate the bi-temporal features using the initial results and swapped features from each temporal branch and To exchange information, we can get enhanced dual-time features and , the enhanced dual-time feature and Contains rich contextual information and temporal relevance.

[0029] As a possible implementation of this embodiment, step S6, fusing the super-neighborhood context attention module, using three super-neighborhood context attention modules to iteratively merge the bi-temporal neighboring domain context features and the multi-scale features to establish time correlation, includes: The super-neighborhood context attention module is integrated, and three super-neighborhood context attention modules are used to iteratively merge the bi-temporal neighboring domain context features. and multi-scale features ; This alignment bridges the domain gap and strengthens the dual-branch semantic relevance, leveraging the complementary strengths of local and wide-area information to enhance global semantic relevance, enabling accurate localization of all varying objects; Based on the merged dual temporal adjacent domain context features and multi-scale features , establish time correlation.

[0030] As a possible implementation of this embodiment, the super-neighborhood context-aware Transformer network includes: Super-Neighborhood Context Branch (HCB) structure, which is used to aggregate extensive context information, overcome the limitation of cropping window and expand the receptive field of the network; Super Neighborhood Context-Aware Position Embedding (HCPE) is used to adaptively integrate neighborhood information, adaptively determine the relative optimal relative position weight encoding, determine the change scale factor by analyzing the dual-phase image, and capture the change scale of the core sample; The Hyper-Neighborhood Context Transformer (HCT) structure is used to fuse multi-scale features and neighborhood context features. It is built on the multi-head attention mechanism introduced by the visual transformer to capture long-range semantic correlations.

[0031] The super-neighborhood context branch structure (HCB) is the basis of the entire HCTNet and has the following functions: used to overcome the limitations of the cropping window and expand the network's receptive field; by aggregating contextual information from a larger range of super-neighborhood images, the ability to crop the surrounding environment of the image is improved; enabling the network to capture a more complete image, significantly enhancing the ability to detect overall scene changes; downsampling, reducing the image resolution to a quarter of the original while retaining key environmental information; using a dual convolutional neural network encoder with a RESNET architecture to gradually capture and integrate comprehensive information about distant neighborhood contexts; the encoder has four layers, each containing two convolutional blocks, each convolutional block containing a 3×3 convolutional layer, a layer and an activation function.

[0032] The Hyper-Neighborhood Context-Aware Location Embedding (HCPE) has the following features: Block to extract and fuse dual-time image features , get the double time difference feature ; Expand the dual time difference feature in the spatial dimension to generate a two-dimensional feature map ; Input the two-dimensional mapping features into the MLP to learn the impact of spatial layout changes and assign different degrees of influence according to the distribution of features; expand its spatial dimension to obtain the same as the dual time difference feature Spatially distributed attention maps with the same spatial scale ;pass and Multiplication can simultaneously consider the impact of spatial dimension and feature dimension on the change scale, and use the conventional change scale prediction method to obtain the change scale factor ; The super-neighborhood context-aware position embedding model prioritizes the core image information and reduces the influence of the surrounding neighborhood context information as the relative distance increases; Gaussian distribution represents the super-neighborhood context image With Core Image The relative mapping relationship of the initial Gaussian distribution is determined by the scale change factor OK, set to mean square error, Chebyshev distance As a weight decay factor; introduce learning parameters , through network learning to autonomously control the weight decay rate, we can obtain the position encoding of super-neighborhood perception .

[0033] Among them, the calculation methods of dual-time difference features, spatial distribution attention map and scale change factor are as follows: Dual time difference features: (1) In the formula, is the dual time difference feature, represent Modules, is the characteristic of dual-time image; Spatial distribution attention map: (2) In the formula, is the spatial distribution attention map, The function represents the expansion of the dimension. represents a multilayer perceptron, The function represents dimensionality reduction; Scale change factor: (3) In the formula, represents the change scale factor, The function is to make predictions, stands for average pooling.

[0034] Among them, the Gaussian distribution and super-neighborhood-aware position coding are calculated according to formulas (4)-(5): (4) In the formula, represents the Chebyshev distance, is the change scale factor; (5) In the formula, To learn the parameters.

[0035] The Hyper-Neighborhood Contextual Transformer (HCT) architecture has the following features: stacking hyper-neighborhood contextual attention modules to capture the generalized contextual features of core samples With local features The remote relationship between Project to In the above, global context information is used to enhance local features; context keywords Sum Generalized context features from core samples , guiding the model to learn extensive contextual information; the features generated by the module become new ,and Keep unchanged, local query Also remain unchanged; calculate the attention weights, obtain the key area information of the image, and gradually strengthen the understanding and fusion of extensive and local features; Transformer composed of two weight-sharing HCAMs and one fusion HCAM; exchange the dual-temporal enhancement features after two iterations and , to ensure that the features in each time branch retain their dual-time characteristics; the third HCAM iteration is performed using the initial results from each time branch and the exchanged features, exchanging information by connecting the dual-time features, focusing on features related to position changes; the dual-time enhanced features after iteration and Contains rich contextual information and temporal correlation, and can distinguish changing areas in bi-temporal images and suppress irrelevant features; fused HCAM uses three HCAM iterations to merge bi-temporal generalized contextual features and local features To bridge the domain gap and strengthen the two-branch semantic association ; Leverage the complementary strengths of local and wide-area information to enhance global semantic relevance, provide a comprehensive view of the network, and accurately locate all changing objects.

[0036] Among them, local query The calculation method of attention weight is as follows: (6) In the formula, is a local feature, is the context feature, represents the weight, Represents contextual keywords, Represents a value; (7) Where HCAM stands for contextual attention module, is the activation function, Represents variance.

[0037] Among them, the dual-time enhanced features after the first two weight-sharing HCAM operations are , and the dual-time enhancement characteristics after the third HCAM iteration Calculate according to formula (8)-(9): (8) (9) In the formula, the generalized context feature , local features And double branch semantic association The calculation methods are expressed by equations (10) and (11): (10) In the formula, represents tensor concatenation, (11).

[0038] like Figure 2 As shown, an embodiment of the present invention provides a super-neighborhood perception remote sensing image change detection device, comprising: Data acquisition module, used to acquire high-resolution remote sensing images and extract their multi-scale features; A feature extraction module for processing the super-neighborhood context image using a convolutional neural network encoder to extract neighborhood context features to model context information; A data merging module for embedding super-neighborhood context-aware positions, merging relative position data into neighborhood context features, adaptively adjusting the position weights of super-neighborhood context images through HCTNet, and synchronizing the relationship between multi-scale features and neighborhood context features; The mechanism introduction module is used to introduce the multi-head attention mechanism based on Vision Transformer to capture long-distance semantic correlation; HCAM stacking module, which is used to capture the long-range dependencies between neighborhood context features and multi-scale features by stacking the Hyper-Neighborhood Context Attention Module (HCAM) of the Transformer model; The feature merging module is used to fuse the super-neighborhood contextual attention module. Three super-neighborhood contextual attention modules are used to iteratively merge the bi-temporal neighboring domain contextual features and multi-scale features to establish temporal correlation. The image detection module is used to build a super-neighborhood context-aware Transformer network and use the super-neighborhood context-aware Transformer network to detect the changed areas in the bi-temporal image.

[0039] An electronic device provided by an embodiment of the present invention includes a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the device is running, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of any super-neighborhood perception remote sensing image change detection method as described above.

[0040] Specifically, the above-mentioned memory and processor can be general-purpose memory and processor, which are not specifically limited here. When the processor runs the computer program stored in the memory, the above-mentioned super-neighborhood perception remote sensing image change detection method can be executed.

[0041] Those skilled in the art will appreciate that the structure of the electronic device does not limit the electronic device and may include more or fewer components than shown in the figure, or combine or split certain components, or arrange the components differently.

[0042] In some embodiments, the electronic device may also include a touch screen that can be used to display a graphical user interface (e.g., a startup interface of an application) and receive user operations on the graphical user interface (e.g., startup operations on an application). The specific touch screen may include a display panel and a touch panel. The display panel may be configured in the form of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), etc. The touch panel may collect the user's contact or non-contact operations on or near it, and generate pre-set operation instructions, for example, the user uses any suitable object such as a finger, stylus, or accessories on or near the touch panel. In addition, the touch panel may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and posture, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into information that the processor can process, and then sends it to the processor, and can receive and execute commands from the processor. In addition, the touch panel can be implemented by various types such as resistive, capacitive, infrared and surface acoustic wave, and any technology developed in the future can also be used to implement the touch panel. Further, the touch panel can cover the display panel, and the user can operate on or near the touch panel covered on the display panel according to the graphical user interface displayed on the display panel. After the touch panel detects the operation on or near it, it is transmitted to the processor to determine the user input, and then the processor provides corresponding visual output on the display panel in response to the user input. In addition, the touch panel and the display panel can be implemented as two independent components or integrated.

[0043] Corresponding to the method for starting the above-mentioned application, an embodiment of the present invention further provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned super-neighborhood perception remote sensing image change detection methods are executed.

[0044] The application startup device provided in the embodiment of the present application can be specific hardware on the device or software or firmware installed on the device. The device provided in the embodiment of the present application, its implementation principle and the technical effect produced are the same as those in the aforementioned method embodiment. For the sake of brief description, the parts not mentioned in the device embodiment can refer to the corresponding contents in the aforementioned method embodiment. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiment, and will not be repeated here.

[0045] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0046] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0047] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0048] In addition, each functional module in the embodiments provided in the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0049] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0050] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0051] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A super-neighborhood-aware remote sensing image change detection method, characterized in that: The steps include: Step S1, obtaining high-resolution remote sensing images and extracting their multi-scale features; Step S2, using a convolutional neural network encoder to process the super-neighborhood context image and extract neighborhood context features to model context information; Step S3, embedding the super-neighborhood context-aware position, merging the relative position data into the neighborhood context features, adaptively adjusting the position weight of the super-neighborhood context image through HCTNet, and synchronizing the relationship between the multi-scale features and the neighborhood context features; Step S4, introduce a multi-head attention mechanism based on Vision Transformer to capture long-distance semantic correlation; Step S5, capturing the long-distance dependency between neighborhood context features and multi-scale features by stacking the super-neighborhood context attention module of the Transformer model; Step S6, fusing the super-neighborhood context attention module, using three super-neighborhood context attention modules to iteratively merge the bi-temporal neighboring domain context features and multi-scale features to establish temporal correlation; Step S7, constructing a super-neighborhood context-aware Transformer network based on steps S2 to S6, and using the super-neighborhood context-aware Transformer network to detect changed areas in the bi-temporal image.

2. The super-neighborhood perception remote sensing image change detection method according to claim 1, characterized in that: The step S2 uses a convolutional neural network encoder to process the super-neighborhood context image, extract neighborhood context features, and model context information, including: Downsample the super-neighborhood context image to a quarter of its original resolution while retaining key context information; A dual CNN encoder is used to perceive a wider context and extract neighborhood context features.

3. The super-neighborhood perception remote sensing image change detection method according to claim 1, characterized in that: The step S3 embeds the super-neighborhood context-aware position, merges the relative position data into the neighborhood context feature, adaptively adjusts the position weight of the super-neighborhood context image through HCTNet, and synchronizes the relationship between the multi-scale feature and the neighborhood context feature, including: Extracting and fusing bi-temporal image features from remote sensing images using Conv-BN-ReLU blocks and , obtain the bitemporal difference feature ∈ ; Expand the bi-temporal difference features in the spatial dimension to generate a two-dimensional feature map ∈ ; The two-dimensional feature map Input into MLP, assign different influence levels to the two-dimensional features according to their distribution, and expand their spatial dimensions to obtain the bi-temporal difference features Spatially distributed attention maps of the same size ; Bitemporal difference feature With spatially distributed attention map Multiply them to obtain the change scale factor σ; Based on the change scale factor σ, HCTNet adaptively adjusts the position weight of the super-neighborhood context image and the relationship between multi-scale features and neighborhood context features.

4. The super-neighborhood perception remote sensing image change detection method according to claim 1, characterized in that: The step S4 introduces a multi-head attention mechanism based on Vision Transformer to capture long-distance semantic correlation; Capturing domain context features and multi-scale features , and optionally domain context features Projection to multi-scale features ; Extracting domain context features Contextual keywords and its value , obtain long-distance semantic correlation and use it to guide the Vision Transformer model to learn extensive contextual information.

5. The super-neighborhood perception remote sensing image change detection method according to claim 1, characterized in that: The step S5 captures the long-distance dependency between the neighborhood context features and the multi-scale features by stacking the super-neighborhood context attention module of the Transformer model, including: Swapping bi-temporal features after two iterations of super-neighborhood contextual attention modules and ; Perform a third iteration of the super-neighborhood contextual attention module to concatenate the bi-temporal features using the initial results and swapped features from each temporal branch and To exchange information, we can obtain enhanced dual-time features and , the enhanced dual-time feature and Contains rich contextual information and temporal relevance.

6. The super-neighborhood perception remote sensing image change detection method according to claim 1, characterized in that: The step S6, fusing the super-neighborhood context attention module, uses three super-neighborhood context attention modules to iteratively merge the bi-temporal neighboring domain context features and the multi-scale features to establish temporal correlation, including: The super-neighborhood context attention module is integrated, and three super-neighborhood context attention modules are used to iteratively merge the bi-temporal neighboring domain context features. and multi-scale features ; Based on the merged dual temporal adjacent domain context features and multi-scale features , establish time correlation.

7. The super-neighborhood perception remote sensing image change detection method according to any one of claims 1 to 6, characterized in that: The super-neighborhood context-aware Transformer network includes: Super-neighborhood context branch structure, which is used to aggregate extensive context information, overcome the limitation of cropping window and expand the receptive field of the network; Super-neighborhood context-aware position embedding structure is used for adaptive neighborhood information integration, adaptively determining the relative optimal relative position weight encoding, determining the change scale factor by analyzing the dual-phase image, and capturing the change scale of the core sample; The Super Neighborhood Context Transformer structure is used to fuse multi-scale features and neighborhood context features, building on the multi-head attention mechanism introduced by the visual transformer to capture long-range semantic correlations.

8. A remote sensing image change detection device based on super-neighborhood perception, characterized in that: include: Data acquisition module, used to acquire high-resolution remote sensing images and extract their multi-scale features; A feature extraction module for processing the super-neighborhood context image using a convolutional neural network encoder to extract neighborhood context features to model context information; A data merging module for embedding super-neighborhood context-aware positions, merging relative position data into neighborhood context features, adaptively adjusting the position weights of super-neighborhood context images through HCTNet, and synchronizing the relationship between multi-scale features and neighborhood context features; The mechanism introduction module is used to introduce the multi-head attention mechanism based on Vision Transformer to capture long-distance semantic correlation; The HCAM stacking module is used to capture the long-range dependencies between neighborhood context features and multi-scale features by stacking the super-neighborhood context attention modules of the Transformer model; The feature merging module is used to fuse the super-neighborhood contextual attention module. Three super-neighborhood contextual attention modules are used to iteratively merge the bi-temporal neighboring domain contextual features and multi-scale features to establish temporal correlation. The image detection module is used to build a super-neighborhood context-aware Transformer network and use the super-neighborhood context-aware Transformer network to detect the changed areas in the bi-temporal image.

9. An electronic device, characterized in that: It includes a processor, a memory and a bus, the memory stores machine-readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of the super-neighborhood perception remote sensing image change detection method as described in any one of claims 1-7.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, executes the steps of the super-neighborhood perception remote sensing image change detection method as described in any one of claims 1 to 7.

Images