A remote sensing image change detection method and related equipment

Abstract

The present invention relates to the field of network detection technology, and discloses a remote sensing image change detection method and related equipment. The method includes acquiring a remote sensing image; establishing a remote sensing image change detection model; and inputting the acquired remote sensing image into the remote sensing image change detection model to perform remote sensing image change detection based on a spatiotemporal difference capture network. The present invention highlights the changed area through a spatial difference capture module, so that the network itself pays more attention to the changed samples. At the same time, the present invention introduces a temporal interactive attention mechanism to further explore the interaction between dual-temporal features, which helps to extract more discriminative features. The present invention achieves simultaneous attention to local details and long-distance context through specific labeling and window limitation strategies at a low cost.

Description

Technical Field

[0001] The present invention relates to the field of network detection technology, and in particular to a remote sensing image change detection method and related equipment. Background Art

[0002] Most existing change detection networks first use a two-stream feature extraction network to extract local features from bitemporal remote sensing images. They then perform feature fusion and enhancement on the bitemporal image pairs. Finally, the obtained multi-scale features are gradually fused from deep to shallow layers to generate classification features, which are then input into a classifier to produce the final change segmentation map. However, these methods do not adequately address the class imbalance issue in change detection. Furthermore, most existing methods directly fuse bitemporal features, neglecting to exploit the potential information hidden in the interaction between the bitemporal features.

[0003] In change detection tasks, the changing regions of interest in a bi-temporal image pair often occupy only a small portion of the image, while the rest remains unchanged. This leads to a serious class imbalance in change detection tasks, significantly impacting model performance. Secondly, the feature interactions between bi-temporal remote sensing images contain potential change information, which can help uncover spatiotemporal cues and strengthen the feature representation of changing regions. However, this point has not received due attention, reducing the network's ability to understand spatiotemporal correlations and suppress irrelevant interference.

[0004] Existing technologies addressing class imbalance primarily focus on loss functions, but this neglects the network's inherent focus on varying samples. Regarding the interaction of bi-temporal features, some methods attempt to achieve interaction by exchanging channel information between them. However, such interactions are incomplete and cannot fully capture variation. Summary of the Invention

[0005] In order to overcome the defects of the above-mentioned prior art, the purpose of the present invention is to provide a remote sensing image change detection method and related equipment to solve the technical problem that the prior art cannot solve the class imbalance and causes the network to tend to classify difficult samples into invariant classes.

[0006] The present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a remote sensing image change detection method, comprising:

[0008] Acquisition of remote sensing images;

[0009] Establish a remote sensing image change detection model;

[0010] The acquired remote sensing images are input into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network.

[0011] Preferably, in the step of establishing a remote sensing image change detection model, the remote sensing image change detection model includes a feature extraction backbone, intra-scale feature enhancement, progressive cross-scale fusion, and a classifier;

[0012] The feature extraction backbone model is ResNet18; the intra-scale feature enhancement model includes spatial difference extraction (SDE) and window-based feature interaction (WFI); the progressive cross-scale fusion includes multiple groups of bilinear interpolation and 1×1 convolutional layers; and the classifier includes two groups of 3×3 convolutional layers.

[0013] Preferably, the acquired remote sensing image is input into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network. The specific process is as follows:

[0014] The remote sensing image is processed through the feature extraction backbone to obtain a local feature map;

[0015] The enhanced features are obtained by capturing the spatiotemporal differences of the local feature maps;

[0016] The enhanced features are fused to obtain the final features;

[0017] Classification is performed based on the final features to generate a change map and complete remote sensing image change detection.

[0018] Furthermore, in the step of capturing the spatiotemporal differences of the local feature map to obtain enhanced features, the specific process is as follows:

[0019] Perform spatial difference extraction (SDE) on the local feature maps in sequence to obtain enhanced features;

[0020] The local feature map and the enhanced feature are obtained through window-based feature interaction (WFI) to obtain the enhanced feature.

[0021] Furthermore, in the step of performing spatial difference extraction (SDE) on the local feature maps to obtain enhanced features, the specific process is as follows:

[0022] The local feature map is passed through the spatial difference attention mechanism SDA to show the change clues hidden in the local features. The spatial difference attention mechanism SDA includes pixel-by-pixel subtraction operation, 3×3 depth-separable convolution layer and 1×1 convolution layer with sigmoid activation function. The expression of the spatial difference attention mechanism SDA is as follows:

[0023]

[0024] in, and Represents a local feature map; Represents a difference feature map; Represents the spatial difference attention map SDA map; 、 、 、 、 They represent absolute functions, Depthwise Separable Convolution, Convolutional layer and sigmoid activation function;

[0025] Based on the spatial difference attention mechanism SDA, the change clues displayed in the local features are subjected to spatial difference extraction SDE to obtain enhanced features. The spatial difference extraction SDE includes an SDA branch and an information mining branch. The SDA branch generates a difference feature map. The information mining branch uses two 3×3 convolutional layers to combine the generated difference feature map with the change clues hidden in the local features to obtain enhanced features. The expression of the spatial difference extraction SDE is as follows:

[0026]

[0027] in, Indicates the preliminary fusion features of the two temporal states; and Represent channel-by-channel connections and convolution; To indicate enhanced features.

[0028] Furthermore, the local feature map is combined with the enhanced feature through the window-based feature interaction (WFI) to obtain the enhanced feature. The specific process is as follows:

[0029] The local feature map and enhanced features are reshaped into sequences through LayerNorm and Linear layers and mapped to obtain Query ( )、Key( ) and Value ( ) token, and Query ( )、Key( ) and Value ( ) token input to the window-based temporal interaction attention layer W-TIA is divided into feature blocks and , the window size is a×a; based on the window being divided into n subspaces along the channel dimension, the expression is as follows:

[0030]

[0031] in, Indicates the number of windows, represents the number of subspaces, 、 and The size of ; Represents the fusion feature block;

[0032] Feature block based and Get the interaction and fusion features of information in each fixed window , the expression is as follows:

[0033]

[0034] in, represents the scale factor, Represents learnable parameters containing relative position cues;

[0035] The interaction and fusion features of the obtained information are combined according to the location information Combined into the first intermediate feature , this feature is rich in global relationship and time information, and at the same time, it strengthens the feature Add to first intermediate feature In the article, we enrich the content from the perspective of local details. , generate the second intermediate feature ;Will Apply the second linear layer LayerNorm layer and the multi-layer perceptron layer MLP layer, combine the residual connection and reshape operation to generate enhanced features .

[0036] In a second aspect, the present invention further provides a remote sensing image change detection system, comprising:

[0037] Image acquisition module, used to acquire remote sensing images;

[0038] Model building module, used to build remote sensing image change detection model;

[0039] The image change detection module is used to input the acquired remote sensing image into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network.

[0040] Preferably, the image change detection module includes a feature extraction backbone module, an intra-scale feature enhancement module, a progressive cross-scale fusion module, and a classification module;

[0041] The feature extraction backbone module is used to process the remote sensing image through the feature extraction backbone to obtain a local feature map;

[0042] The intra-scale feature enhancement module is used to capture the temporal and spatial differences of local feature maps to obtain enhanced features;

[0043] The progressive cross-scale fusion module is used to fuse the enhanced features to obtain the final features;

[0044] The classification module is used to generate a change map based on the final features and complete the remote sensing image change detection.

[0045] In a third aspect, the present invention also provides a mobile terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the remote sensing image change detection method as described above when executing the computer program.

[0046] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the remote sensing image change detection method as described above are implemented.

[0047] Compared with the prior art, the present invention has the following beneficial technical effects:

[0048] The present invention provides a remote sensing image change detection method. By establishing a remote sensing image change detection model and using a spatiotemporal difference capture network to perform remote sensing image change detection, a large amount of remote sensing image data can be efficiently processed and analyzed. The spatiotemporal difference capture network can accurately capture the temporal and spatial change features in the image, thereby improving the accuracy of change detection. The present invention highlights the change area through the spatial difference capture module, so that the network itself pays more attention to the change samples. At the same time, the present invention introduces a temporal interactive attention mechanism to further explore the interaction between dual temporal features, which helps to extract more discriminative features. The present invention achieves simultaneous attention to local details and long-distance context through specific labeling and window limitation strategies at a low cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Flowchart of a remote sensing image change detection method according to an embodiment of the present invention;

[0050] Figure 2 Flowchart of remote sensing image change detection based on spatiotemporal difference capture network in an embodiment of the present invention;

[0051] Figure 3 This is a diagram of the SDCNet framework in an embodiment of the present invention;

[0052] Figure 4 This is a schematic diagram of the spatial difference attention mechanism SDA in an embodiment of the present invention;

[0053] Figure 5 This is a schematic diagram of the SDE structure for spatial difference extraction in an embodiment of the present invention;

[0054] Figure 6 Schematic diagram of the window-based feature interaction WFI structure in an embodiment of the present invention;

[0055] Figure 7 Schematic diagram of the W-TIA structure in an embodiment of the present invention;

[0056] Figure 8 Schematic diagram of the remote sensing image change detection system according to an embodiment of the present invention;

[0057] Figure 9 2. This is a schematic diagram of the internal structure of an image change detection module according to an embodiment of the present invention;

[0058] In the figure: 1. Image acquisition module; 2. Model building module; 3. Image change detection module; 31. Feature extraction backbone module; 32. Intra-scale feature enhancement module; 33. Progressive cross-scale fusion module; 34. Classification module. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0060] The purpose of the present invention is to provide a remote sensing image change detection method and related equipment to solve the technical problem that the existing technology cannot solve the class imbalance and causes the network to tend to classify difficult samples into unchanged classes.

[0061] The present invention is described in further detail below with reference to the accompanying drawings:

[0062] See also Figure 1 In one embodiment of the present invention, a remote sensing image change detection method is provided, comprising the following steps:

[0063] Step 1, acquiring remote sensing images;

[0064] Step 2: Establish a remote sensing image change detection model;

[0065] Specifically, according to Figure 3 As shown, the remote sensing image change detection model includes a feature extraction backbone, intra-scale feature enhancement, progressive cross-scale fusion, and a classifier;

[0066] The feature extraction backbone model is ResNet18; the intra-scale feature enhancement model includes spatial difference extraction (SDE) and window-based feature interaction (WFI); the progressive cross-scale fusion includes multiple groups of bilinear interpolation and 1×1 convolutional layers; and the classifier includes two groups of 3×3 convolutional layers.

[0067] Step 3: Input the acquired remote sensing image into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network.

[0068] Specifically, according to Figure 2 As shown, the specific process is as follows:

[0069] S1, the remote sensing image is processed through the feature extraction backbone to obtain the local feature map;

[0070] S2, captures the spatiotemporal differences of local feature maps to obtain enhanced features;

[0071] Specifically, the specific process is as follows:

[0072] S21, perform spatial difference extraction (SDE) on the local feature map in sequence to obtain enhanced features;

[0073] The specific process is as follows:

[0074] The local feature map is passed through the spatial difference attention mechanism SDA to reveal the change clues hidden in the local features, where the spatial difference attention mechanism SDA includes pixel-by-pixel subtraction operation, 3×3 depth-wise separable convolution layer and 1×1 convolution layer with sigmoid activation function.

[0075] according to Figure 4 As shown, when a pair of local feature maps are input When the spatial difference attention mechanism SDA first applies a pixel-by-pixel subtraction operation to obtain the difference feature map This step aims to initially reveal the changes between the dual-phase remote sensing images. Then, a progressive refinement strategy is used to further explore the clues of changes. First, we introduce Depthwise separable convolution will Map to .choose The purpose of depthwise separable convolution is to effectively control the growth of the number of parameters while maintaining a large receptive field. The convolutional layer will Convert to SDA map Here, the spatial difference attention map SDA map can be regarded as a mask, in which the value ranges from 0 to 1, indicating the degree of attention the network needs to pay to these areas. The expression of the spatial difference attention mechanism SDA is as follows:

[0076]

[0077] in, and Represents a local feature map; Represents a difference feature map; Represents the spatial difference attention map SDA map; 、 、 、 、 They represent absolute functions, Depthwise Separable Convolution, Convolutional layer and sigmoid activation function;

[0078] In this embodiment, the spatial difference attention mechanism (SDA) ensures that the SDCNet can more accurately locate regions of change, as it can deemphasize or highlight regions of false or true change. Furthermore, because the spatial difference attention mechanism (SDA) assigns higher weights to regions of true change, it can place greater emphasis on the limited number of positive examples when training the SDCNet. This approach can mitigate the negative effects of the imbalance problem.

[0079] Based on the spatial difference attention mechanism SDA, the change clues displayed in the local features are extracted by spatial difference SDE to obtain enhanced features, such as Figure 5 As shown in the figure, the spatial difference extraction SDE includes an SDA branch and an information mining branch. The SDA branch generates a difference feature map; the information mining branch uses two 3×3 convolutional layers to combine the generated difference feature map with the change clues hidden in the local features to obtain enhanced features.

[0080] In this embodiment, the two feature maps are connected in the channel dimension, and the first Convolution Generation , which contains detailed clues in the dual-time remote sensing imagery. Then, and Connect together to ensure that basic change information can be integrated. After convolution, we can generate , which contains the detail change information hidden in the two RS images. Next, the SDA map will with Multiply to get the final output This step can not only suppress the The irrelevant clues in can further strengthen the representation of the changed area. Possessing rich knowledge, it can help the model generate perfect change maps.

[0081] Among them, the expression of spatial difference extraction SDE is as follows:

[0082]

[0083] in, Indicates the preliminary fusion features of the two temporal states; and Represent channel-by-channel connections and convolution; To indicate enhanced features.

[0084] In this embodiment, spatial difference extraction can effectively guide SDCNet to focus on the changed areas and reduce the negative impact of the imbalance problem.

[0085] S22, the local feature map and the enhanced feature are obtained through the window-based feature interaction WFI to obtain the enhanced feature.

[0086] The specific process is as follows:

[0087] The local feature map and enhanced features are reshaped into sequences through LayerNorm and Linear layers and mapped to obtain Query ( )、Key( ) and Value ( ) token, and Query ( )、Key( ) and Value ( ) token input to the window-based temporal interaction attention layer W-TIA is divided into feature blocks and , the window size is a×a; based on the window, it is divided into n subspaces along the channel dimension.

[0088] This example introduces the self-attention mechanism Transformer. However, when the Transformer is directly applied to When the self-attention mechanism is used, the detailed information will be ignored or blurred. Therefore, the window version of Transformer is adopted. At the same time, considering the importance of analyzing the two time information for the RSCD task, in addition to In addition, local feature maps should also be considered and .

[0089] In this embodiment, the window-based feature interaction block WFI has a structure as follows: Figure 6It consists of a LayerNorm layer with a linear layer, a window-based temporal interaction attention (W-TIA) layer, a LayerNorm layer, and a multi-layer perceptron (MLP) layer (composed of two linear maps). They are connected in a residual manner.

[0090] When the general When input into the WFI module, they will be reshaped into sequences by the LayerNorm and Linear layers and mapped to the Query ( )、Key( ) and Value ( ) token. Then, they are sent to the W-TIA layer, which is the core of WFI, as shown in the diagram Figure 7 It is essentially a simplified version of the Swim Transformer. It uses a fixed meshing scheme instead of a shifting window to transform and Divided into various feature blocks, where the window size is Instead of exchanging information by moving windows, we only need to focus on the pixels that have changed within the window. This approach can also reduce the parameter size of the entire model. At the same time, they divide the channel dimension into n subspaces to help WFI capture contextual information in terms of the channel. The expression is as follows:

[0091]

[0092] in, Indicates the number of windows, represents the number of subspaces, 、 and The size of ; Represents the fusion feature block;

[0093] Feature block based and Get the interaction and fusion features of information in each fixed window , the expression is as follows:

[0094]

[0095] in, represents the scale factor, Represents learnable parameters containing relative position cues;

[0096] In this embodiment, on the one hand, and The operation of realizes the interaction of time knowledge and helps WFI discover hidden change information. On the other hand, due to the use of ,The changed regions determined by SDE can also be fused, which helps WFI identify the changed regions more accurately based on ,temporal clue analysis.

[0097] The interaction and fusion features of the obtained information are combined according to the location information Combined into the first intermediate feature , the location information is rich in global relationship and time information, and at the same time, the features are strengthened Add to first intermediate feature In the article, we enrich the content from the perspective of local details. , generate the second intermediate feature ;Will Apply the second linear layer LayerNorm layer and the multi-layer perceptron layer MLP layer, combine the residual connection and reshape operation to generate enhanced features .

[0098] In this embodiment, window-based feature interaction can focus on mining the interactions between bi-temporal features and deeply explore spatiotemporal clues to obtain more robust feature representation.

[0099] S3, the enhanced features are fused to obtain the final features;

[0100] S4, classify and generate change maps based on the final features to complete remote sensing image change detection.

[0101] This embodiment first proposes a spatial difference attention mechanism and then a spatial difference capture module. This mechanism can highlight the changing regions in a bi-temporal image pair, guiding the network to focus more on the changing targets, thereby improving the accuracy of change detection. The window-based feature interaction module designed in this invention uses temporal interaction attention to deeply explore the interactions between bi-temporal features, enhancing the discriminability of spatiotemporal cues and obtaining richer feature representations.

[0102] In summary, the remote sensing image change detection method provided by this embodiment, compared to most methods that alleviate the class imbalance problem by weighting the loss function, the present invention highlights the change area through the spatial difference capture module, so that the network itself pays more attention to the positive samples of change during the training process. Secondly, compared to other methods that simply exchange bi-temporal features, the present invention introduces a temporal interactive attention mechanism to further explore the interaction between bi-temporal features, which helps to extract more discriminative features. Finally, compared to other common Transformer and CNN combination methods in change detection tasks, the present invention achieves simultaneous attention to long-range context and local details through specific labeling (implemented by SDE) and intra-window spatiotemporal exploration (implemented by WFI) strategies at a low cost.

[0103] Example 2

[0104] according to Figure 8 As shown, this embodiment provides a remote sensing image change detection system, including:

[0105] Image acquisition module 1, used for acquiring remote sensing images;

[0106] Model building module 2, used to build a remote sensing image change detection model;

[0107] The image change detection module 3 is used to input the acquired remote sensing image into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network.

[0108] Specifically, according to Figure 9 As shown, the image change detection module includes a feature extraction backbone module 31, an intra-scale feature enhancement module 32, a progressive cross-scale fusion module 33, and a classification module 34;

[0109] The feature extraction backbone module 31 is used to process the remote sensing image through the feature extraction backbone to obtain a local feature map;

[0110] The intra-scale feature enhancement module 32 is used to capture the spatiotemporal differences of the local feature map to obtain enhanced features;

[0111] A progressive cross-scale fusion module 33 is used to fuse the obtained enhanced features to obtain the final features;

[0112] The classification module 34 is used to perform classification based on the final features to generate a change map and complete remote sensing image change detection.

[0113] In the actual application process of this embodiment, the specific operations are as follows:

[0114] When a pair of remote sensing images When inputting SDCNet, the feature extraction backbone module 31 is first used to learn local information from it. Here, and Respectively represent the height, weight and channel of the remote sensing image. Thanks to the special structure of the feature extraction backbone ResNet18, four pairs of local feature maps corresponding to different scales can be obtained, which are recorded as ,in represents the level of feature maps. The size of these feature maps is ,in , For convenience, The local features are then fed into the intra-scale feature enhancement module 32, which aims to explore temporal cues and improve features at different single scales. After this module, we can obtain enhanced features in which the changed areas are highly focused and have rich representations. Next, we apply the progressive cross-scale fusion module 33 to fuse them into , which can take into account the relationship between different scales, and then detect the hidden change areas in the RS image from coarse to fine. , we use the classification module 34 to generate the change map .

[0115] Example 3

[0116] The present invention also provides a mobile terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, such as a remote sensing image change detection management program.

[0117] When the processor executes the computer program, the steps of the remote sensing image change detection method are implemented, for example:

[0118] Acquisition of remote sensing images;

[0119] Establish a remote sensing image change detection model;

[0120] The acquired remote sensing images are input into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network.

[0121] Among them, the acquired remote sensing image is input into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network. The specific process is as follows:

[0122] The remote sensing image is processed through the feature extraction backbone to obtain a local feature map;

[0123] The enhanced features are obtained by capturing the spatiotemporal differences of the local feature maps;

[0124] The enhanced features are fused to obtain the final features;

[0125] Classification is performed based on the final features to generate a change map and complete remote sensing image change detection.

[0126] Alternatively, when the processor executes the computer program, the functions of each module in the above system are realized, for example:

[0127] Image acquisition module 1, used for acquiring remote sensing images;

[0128] Model building module 2, used to build a remote sensing image change detection model;

[0129] The image change detection module 3 is used to input the acquired remote sensing image into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network.

[0130] The image change detection module includes a feature extraction backbone module 31, an intra-scale feature enhancement module 32, a progressive cross-scale fusion module 33, and a classification module 34;

[0131] The feature extraction backbone module 31 is used to process the remote sensing image through the feature extraction backbone to obtain a local feature map;

[0132] The intra-scale feature enhancement module 32 is used to capture the spatiotemporal differences of the local feature map to obtain enhanced features;

[0133] A progressive cross-scale fusion module 33 is used to fuse the obtained enhanced features to obtain the final features;

[0134] The classification module 34 is used to perform classification based on the final features to generate a change map and complete remote sensing image change detection.

[0135] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program in the mobile terminal.

[0136] For example, the computer program may be divided into an image acquisition module 1, a model building module 2, and an image change detection module 3; wherein the image change detection module 33 includes a feature extraction backbone module 31, an intra-scale feature enhancement module 32, a progressive cross-scale fusion module 33, and a classification module 34;

[0137] The specific functions of each module are as follows:

[0138] Image acquisition module 1, used for acquiring remote sensing images;

[0139] Model building module 2, used to build a remote sensing image change detection model;

[0140] The image change detection module 3 is used to input the acquired remote sensing image into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network.

[0141] The feature extraction backbone module 31 is used to process the remote sensing image through the feature extraction backbone to obtain a local feature map;

[0142] The intra-scale feature enhancement module 32 is used to capture the spatiotemporal differences of the local feature map to obtain enhanced features;

[0143] A progressive cross-scale fusion module 33 is used to fuse the obtained enhanced features to obtain the final features;

[0144] The classification module 34 is used to perform classification based on the final features to generate a change map and complete remote sensing image change detection.

[0145] The mobile terminal may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The mobile terminal may include, but is not limited to, a processor and a memory.

[0146] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the mobile terminal, connecting various parts of the entire mobile terminal using various interfaces and lines.

[0147] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the mobile terminal by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0148] The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as sound playback or image playback); the data storage area may store data generated based on the use of the mobile phone (such as audio data and a phone book). Furthermore, the memory may include high-speed random access memory (RAM) and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0149] Example 4

[0150] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the remote sensing image change detection method are implemented.

[0151] If the module / unit integrated in the mobile terminal is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0152] Based on this understanding, the present invention can implement all or part of the processes in the above-mentioned method by means of a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above-mentioned method for scheduling aggregated reinforcement learning resources. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form.

[0153] The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0154] It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the 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 by the scope of protection of the claims of the present invention.

Claims

1. A remote sensing image change detection method, characterized in that: include: Acquisition of remote sensing images; Establish a remote sensing image change detection model; The acquired remote sensing images are input into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network; The specific process of inputting the acquired remote sensing image into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network is as follows: The remote sensing image is processed through the feature extraction backbone to obtain a local feature map; The enhanced features are obtained by capturing the spatiotemporal differences of the local feature maps; The enhanced features are fused to obtain the final features; Classify and generate change maps based on the final features to complete remote sensing image change detection; Among them, in the step of capturing the spatiotemporal difference of the local feature map to obtain enhanced features, the specific process is as follows: Perform spatial difference extraction (SDE) on the local feature maps in sequence to obtain enhanced features; The local feature map and the enhanced feature are obtained through the window-based feature interaction WFI to obtain the enhanced feature; Among them, in the step of sequentially performing spatial difference extraction (SDE) on the local feature map to output enhanced features, the specific process is as follows: The local feature map is passed through the spatial difference attention mechanism SDA to show the change clues hidden in the local features. The spatial difference attention mechanism SDA includes pixel-by-pixel subtraction operation, 3×3 depth-separable convolution layer and 1×1 convolution layer with sigmoid activation function. The expression of the spatial difference attention mechanism SDA is as follows: in, and Represents a local feature map; Represents a difference feature map; Represents the spatial difference attention map SDA map; 、 、 、 、 They represent absolute functions, Depthwise Separable Convolution, Convolutional layer and sigmoid activation function; Based on the spatial difference attention mechanism SDA, the change clues displayed in the local features are subjected to spatial difference extraction SDE to obtain enhanced features. The spatial difference extraction SDE includes an SDA branch and an information mining branch. The SDA branch generates a difference feature map. The information mining branch uses two 3×3 convolutional layers to combine the generated difference feature map with the change clues hidden in the local features to obtain enhanced features. The expression of the spatial difference extraction SDE is as follows: in, Indicates the preliminary fusion features of the two temporal states; and Represent channel-by-channel connections and convolution; To indicate enhanced characteristics; The specific process of combining the local feature map with the enhanced feature through the window-based feature interaction (WFI) to obtain the enhanced feature is as follows: The local feature map and enhanced features are reshaped into sequences through LayerNorm and Linear layers and mapped to obtain Query ( )、Key( ) and Value ( ) token, and Query ( )、Key( ) and Value ( ) token input to the window-based temporal interaction attention layer W-TIA is divided into feature blocks and , the window size is a×a; based on the window being divided into n subspaces along the channel dimension, the expression is as follows: in, Indicates the number of windows, represents the number of subspaces, 、 and The size of ; Represents the fusion feature block; Feature block based and Get the interaction and fusion features of information in each fixed window , the expression is as follows: in, represents the scale factor, Represents learnable parameters containing relative position cues; The interaction and fusion features of the obtained information are combined according to the location information Combined into the first intermediate feature , this feature is rich in global relationship and time information, and at the same time, it strengthens the feature Add to first intermediate feature In the article, we enrich the content from the perspective of local details. , generate the second intermediate feature ;Will Apply the second linear layer LayerNorm layer and the multi-layer perceptron layer MLP layer, combine the residual connection and reshape operation to generate enhanced features .

2. The remote sensing image change detection method according to claim 1, characterized in that: In the step of establishing a remote sensing image change detection model, the remote sensing image change detection model includes a feature extraction backbone, intra-scale feature enhancement, progressive cross-scale fusion, and a classifier; The feature extraction backbone model is ResNet18; the intra-scale feature enhancement model includes spatial difference extraction (SDE) and window-based feature interaction (WFI); the progressive cross-scale fusion includes multiple groups of bilinear interpolation and 1×1 convolutional layers; and the classifier includes two groups of 3×3 convolutional layers.

3. A remote sensing image change detection system, based on a remote sensing image change detection method according to any one of claims 1 to 2, characterized in that: include: Image acquisition module, used to acquire remote sensing images; Model building module, used to build remote sensing image change detection model; The image change detection module is used to input the acquired remote sensing image into the remote sensing image change detection model to perform remote sensing image change detection based on the spatiotemporal difference capture network.

4. A remote sensing image change detection system according to claim 3, characterized in that: The image change detection module includes a feature extraction backbone module, an intra-scale feature enhancement module, a progressive cross-scale fusion module and a classification module; The feature extraction backbone module is used to process the remote sensing image through the feature extraction backbone to obtain a local feature map; The intra-scale feature enhancement module is used to capture the temporal and spatial differences of local feature maps to obtain enhanced features; The progressive cross-scale fusion module is used to fuse the enhanced features to obtain the final features; The classification module is used to generate a change map based on the final features and complete the remote sensing image change detection.

5. A mobile terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the remote sensing image change detection method according to any one of claims 1 to 2 are implemented.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the remote sensing image change detection method according to any one of claims 1 to 2 are implemented.

Images