Remote sensing change detection method based on deep learning based on hybrid spatial-frequency experts

Abstract

The present invention relates to a deep learning remote sensing change detection method based on hybrid space-frequency experts, comprising: selecting a twin ConvNeXt network as a spatial domain expert to extract the coarse-fine spatial domain local difference features of dual-temporal images; selecting a twin discrete wavelet transform as a frequency domain expert to extract the global difference features of the high-low frequency domain of the dual-temporal images; aggregating the local-global space-frequency features of the corresponding scales to construct a multi-scale hybrid space-frequency monomer feature; designing a random hybrid space-frequency monomer fusion module to randomly select different space-frequency features for fusion and enhance complementarity, and activate effective fusion based on contribution weights; randomly selecting hybrid space-frequency monomers to form a group and perform probabilistic weighting to adaptively select the optimal group; and constructing a lightweight decoder to gradually restore the enhanced multi-scale space-frequency features to the resolution of the original image. This application achieves complementary optimization of cross-domain features, reduces invalid parameters, and improves the accuracy and speed of remote sensing change detection.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing change detection, and in particular to a remote sensing change detection method based on deep learning of hybrid space-frequency experts. Background Art

[0002] Change detection is a key application in remote sensing. Remote sensing technology can accurately capture dynamic temporal and spatial changes in ground features over a large scale, including changes in buildings, water bodies, cultivated land, and roads. This is crucial for post-disaster assessments and environmental monitoring.

[0003] With the development of deep learning, deep neural networks have gradually become the mainstream method for change detection. Compared with traditional methods such as image difference, deep neural networks can automatically learn high-dimensional abstract features of spatial and temporal dimensions in remote sensing images, which is conducive to improving the accuracy and robustness of change detection.

[0004] The current mainstream methods for change detection based on deep neural networks include the following two:

[0005] (1) Deep neural network based on spatial domain

[0006] These methods primarily rely on architectures such as convolutional neural networks, Transformers, and Mamba to achieve superior visual representation capabilities by extracting deep features in the spatial domain. Remote sensing objects exhibit significant scale effects, and current research typically employs multi-scale feature fusion to further improve the robustness and generalization of remote sensing object extraction. In particular, fine-scale features are beneficial for depicting spatial details, while coarse-scale features are adept at describing semantic information about objects. Therefore, by fusing multi-scale features, the richness of visual representations can be further enhanced, improving visual understanding of objects.

[0007] (2) Deep neural network based on frequency domain

[0008] This type of method uses frequency-domain transformations, such as discrete wavelet transforms, to convert remote sensing image features from the spatial domain to the frequency domain for spatiotemporal information expression. Current research often utilizes multi-spectral feature fusion to further enhance the representation of ground objects. High-frequency features are sensitive to edge information, while low-frequency features are good at capturing structural information. Therefore, fusing multi-frequency features can improve change detection accuracy.

[0009] In summary, current technical research still has shortcomings, such as: 1) Current research is limited to single feature domain representation, and the method of joint cross-domain representation is still imperfect, and cannot fully utilize the respective advantages to complement features; 2) In the cross-domain representation process, as the number of parameters increases, how to balance computational consumption and accuracy improvement is still worth considering; 3) In cross-domain feature combination, how to scientifically measure effective combination and reduce noise is still imperfect. Summary of the Invention

[0010] The present invention provides a deep learning remote sensing change detection method based on hybrid space-frequency experts. It uses expert networks in different feature domains as encoders to extract spatial local features and frequency global features, and then adopts a hybrid expert network for individual and group interactive decoding, thereby complementing the advantages of space-frequency features, enhancing the richness of representation, and reducing the computational consumption of redundant features, thereby improving the accuracy of remote sensing ground object change detection.

[0011] The present invention provides a deep learning remote sensing change detection method based on hybrid space-frequency experts, which adopts the following technical solutions:

[0012] The deep learning remote sensing change detection method based on hybrid space-frequency experts includes the following steps:

[0013] S1, selects the twin ConvNeXt network as the spatial domain expert to extract the multi-scale coarse-fine spatial domain local difference features of the dual-temporal remote sensing image;

[0014] S2, uses twin discrete wavelet transform as the frequency domain expert to extract the high-low frequency domain global difference features of the multi-spectral of the dual-temporal remote sensing image;

[0015] S3, constructs multi-scale mixed space-frequency monomer features by aggregating local-global space-frequency features of corresponding scales;

[0016] S4, designs a random mixed space-frequency monomer fusion module, randomly selects different space-frequency features for fusion, enhances complementarity and calculates contribution weights to activate effective fusion;

[0017] S5, design a random mixed space-frequency group decision module, randomly select different mixed space-frequency monomers to form a group and perform probabilistic weighting to adaptively select the optimal decision group;

[0018] S6, builds a lightweight decoder to gradually restore the enhanced multi-scale space-frequency features to the resolution of the original image and generate change detection results.

[0019] Preferably, said S1 comprises the following steps:

[0020] S11, extracting the deep spatial features of dual-temporal remote sensing images:

[0021] ConvNeXt is selected as the spatial domain expert and as the feature encoder, and a weight sharing method is adopted based on the twin architecture to obtain dual-temporal remote sensing images. and Multi-scale deep visual features in the spatial domain and ,Right now:

[0022] (1)

[0023] S12, aligns the dual-temporal multi-scale coarse-fine spatial domain local difference features, namely:

[0024] (2)

[0025] in, Select the concatenate function for multi-dimensional spatiotemporal alignment. There are four multi-scale spatial local difference features from coarse to fine.

[0026] Preferably, said S2 comprises the following steps:

[0027] S21, using the step-by-step decoupling method to select discrete wavelet transform to extract the multi-spectral frequency domain features of single-temporal remote sensing images, namely:

[0028] (3)

[0029] in, The detailed iterative process of the step-by-step frequency decomposer is:

[0030] (4)

[0031] Among them, DWT is the discrete wavelet transform, where The SwinTransformer block is used to enhance the deep visual representation capability of the global spectrum. Including high-high, high-low, and low-high frequency components decomposed by DWT, including low-low frequency components;

[0032] S22, extracting global deep visual features in the multi-spectral frequency domain of dual-temporal remote sensing images using a twin architecture based on a step-by-step frequency decomposer and ,Right now:

[0033] = (5)

[0034] S23, align the frequency domain global difference features of the dual-phase high-low multi-spectrum, namely:

[0035] (6)

[0036] in, Select the concatenate function as the frequency domain feature alignment function, These are the frequency domain global difference features of multi-spectrum at four scales from high to low.

[0037] Preferably, the S3 is:

[0038] The corresponding multi-scale spatial domain local difference features and multi-spectral frequency domain global difference features are fused through the aggregation process and used as mixed space-frequency monomer features, namely:

[0039] (7)

[0040] in, Select the concatenate function as the aggregation function to aggregate space-frequency features, where is the corresponding mixed space-frequency monomer feature of the output.

[0041] Preferably, the S4 includes a random mixing space-frequency monomer fusion module, and the formula is:

[0042] (8)

[0043] in, is a random mixed space-frequency monomer fusion module, where Indicates the frequency of random sampling and the number of spatial monomer experts, is the ith mixed space-frequency monomer, It is a mixed space feature monomer expert library, is the mixed frequency monomer expert library, To output the result;

[0044] S4 includes the following steps:

[0045] S41, design random optimization strategy;

[0046] Using the sampling strategy without replacement, n spatial monomer experts are randomly selected from S and n frequency monomer experts are randomly selected from F as auxiliary monomer experts to enhance the mixed spatial-frequency monomer features. ;

[0047] S42, fusion of different spatial / frequency monomer experts to enhance the hybrid spatial / frequency monomer features;

[0048] The input mixed space-frequency monomer feature As the monomer expert to be enhanced, for the randomly selected spatial / frequency monomer expert As an auxiliary monomer expert, where i and m are feature index values;

[0049] S421, align the feature dimensions of the auxiliary single expert and the single expert to be enhanced, namely:

[0050] (9)

[0051] in, is the upsampling convolution function, is the downsampling convolution function, is the concatenate function;

[0052] S422, design a hybrid spatial-frequency attention module, based on Mamba attention, to focus on and enhance the complementarity between different scale and spectral features, namely:

[0053] (10)

[0054] in, is the linear projection layer, It's multi-headed attention. is a multilayer perceptron layer, for:

[0055] (11)

[0056] in, is the linear projection layer, is the convolutional layer, is the SiLU activation function, It is a state space model, which uses vertical mode and horizontal mode to scan data;

[0057] S43, sequentially enhancing the mixed space-frequency monomer features based on step S42 for the randomly selected n space monomer experts and n frequency monomer experts;

[0058] S44, design a hybrid space-frequency monomer expert network, calculate the contribution weight of the fusion one by one through the gated network adaptive calculation, activate the effective features, and eliminate the invalid enhancement features, that is:

[0059] (12)

[0060] in, is the output feature, is the softmax function, which is used to calculate the contribution probability. The optimized mixed space-frequency monomer characteristics in step S43 are: As routing and using the TopK algorithm, the optimal feature is selected as output.

[0061] Preferably, the S5 includes a random mixed space-frequency group decision module, the formula of which is:

[0062] (13)

[0063] in, By constructing a mixed space-frequency group To enhance, Represents an enhanced mixed space-frequency monomer library The Features, represents the constructed mixed space-frequency group, n represents the number of constructed groups, and includes the following steps:

[0064] S5 includes the following steps:

[0065] S51, randomly select the optimized mixed space-frequency monomer characteristics to form a mixed space-frequency group, which is:

[0066] (14)

[0067] in , p represents the number of monomers randomly drawn from D, q represents the qth population constructed, i is the feature index, and the selected monomers are upsampled or downsampled with The feature dimensions are aligned, and the concatenate function is used to concatenate the features of all selected individuals to construct n groups;

[0068] S52, based on the constructed mixed space-frequency group, makes an optimal decision, namely:

[0069] (15)

[0070] in, For output, The softmax function is used, l represents the lth group, to calculate the contribution probability of each group decision. As a router and using the TopK algorithm, the optimal group decision result is selected as the output;

[0071] S53, using group decision making for optimized hybrid space-frequency monomers The various elements of the group are enhanced.

[0072] Preferably, the S6 comprises the following steps:

[0073] S61, preset lightweight decoder:

[0074] (16)

[0075] in, Represented as an upsampling convolutional layer, For Cat operation, The Mamba attention in step S422 is used to further enhance the feature expression;

[0076] S62, output results;

[0077] (17)

[0078] in, Represented as a convolutional layer, Used to output classification probabilities.

[0079] In summary, the present invention has the following beneficial technical effects:

[0080] 1. The present invention extracts spatial local features and frequency global features from different domains through an encoder, and adopts a hybrid expert network for individual and group interactive decoding, thereby complementing the advantages of spatial-frequency features, enhancing the richness of representation, reducing the computational consumption of redundant features, and improving the accuracy of remote sensing ground object change detection.

[0081] 2. This invention considers the representation capabilities of deep neural networks for remote sensing change detection from a multi-feature domain and multi-angle perspective, including: multi-scale-multi-spectral, global-local, edge-structure, and detail-semantic. This has the following advantages: 1) ensuring that network optimization focuses on improving specific tasks, reducing mutual interference between the aforementioned complex feature optimizations, thereby improving the modeling and expression capabilities of cross-domain features; 2) increasing the richness of representations across feature domains, improving the ability to perceive spatial-level details and abstract features of remote sensing objects, and the ability to capture frequency-level edge and structural features of remote sensing objects;

[0082] 3. This invention demonstrates the advantages of rationality, robustness, and generalization in the fusion of multi-scale spatial and multi-spectral frequency features. This is demonstrated by: 1) designing a corresponding hybrid expert network to calculate the contribution weights of the fusion between individual and group expert features, enabling a scientific and rational evaluation of the representational capabilities of the fusion method; 2) randomly selecting the fusion method for individual and group expert features, which helps reduce the interference of individual bias and noise, avoids network overfitting, and improves the generalization and robustness of the network. 3) designing and selecting the fusion method for individual and group expert features, by fusing a variety of spatial-frequency combinations of individual features, taking into account the possibility of multiple fusion methods, reducing the error and uncertainty of feature fusion through network optimization, and improving generalization capability.

[0083] 4. This invention considers the balance between computing power consumption and computational accuracy, which is conducive to achieving load balancing. By designing a hybrid expert network of individuals and groups, while fully considering the complementarity, richness, and diversity of the fusion of single-scale spectrum and multi-scale spectrum features, it further enhances the generalization and robustness of the model by activating beneficial features and discarding invalid features, while reducing a large amount of invalid calculations and saving network inference consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 It is a schematic diagram of a flowchart in an embodiment of the present invention.

[0085] Figure 2 Schematic diagram of a random mixing space-frequency monomer fusion module in an embodiment of the present invention.

[0086] Figure 3 Schematic diagram of a hybrid space-frequency attention module according to an embodiment of the present invention.

[0087] Figure 4 Schematic diagram of the scanning method in the hybrid space-frequency attention module in an embodiment of the present invention.

[0088] Figure 5 This is the visual result of the method of the present invention and the comparison network on the WB-CD dataset in the embodiment of the present invention. DETAILED DESCRIPTION

[0089] The present invention will be further described in detail below with reference to the accompanying drawings.

[0090] The embodiment of the present invention discloses a remote sensing change detection method based on deep learning of hybrid space-frequency experts.

[0091] This example uses building change detection as an example and uses the open source remote sensing building change detection dataset WB-CD dataset. The dataset is sampled from two cities, Christchurch and New Zealand, and covers an urban area of ​​20.5 km. 2 , including two time periods: 2012 and 2016, which include a large number of independent buildings, increasing from 12796 to 16077. Its images will be cut into non-overlapping pairs of samples with a size of 256×256 pixels, and the training, validation and test datasets will be generated in a ratio of 1:1:8.

[0092] Reference Figure 1 , the deep learning remote sensing change detection method based on hybrid space-frequency experts includes the following steps:

[0093] S1, using the twin ConvNeXt network as the spatial domain expert to extract the multi-scale coarse-fine spatial domain local difference features of the dual-temporal remote sensing image:

[0094] Specifically, S1 includes the following steps:

[0095] S11, extracting the deep spatial features of dual-temporal remote sensing images:

[0096] ConvNeXt is selected as the spatial domain expert and as the feature encoder, and a weight sharing method is adopted based on the twin architecture to obtain dual-temporal remote sensing images. and Multi-scale deep visual features in the spatial domain and ,Right now:

[0097] (1)

[0098] S12, aligns the dual-temporal multi-scale coarse-fine spatial domain local difference features, namely:

[0099] (2)

[0100] in, Select the concatenate function for multi-dimensional spatiotemporal alignment. There are four multi-scale spatial local difference features from coarse to fine.

[0101] S2 uses the twin discrete wavelet transform as the frequency domain expert to extract the high-low frequency domain global difference features of the multi-spectral of the dual-temporal remote sensing image:

[0102] Specifically, S2 includes the following steps:

[0103] S21, using the step-by-step decoupling method to select discrete wavelet transform to extract the multi-spectral frequency domain features of single-temporal remote sensing images, namely:

[0104] (3)

[0105] in, The detailed iterative process of the step-by-step frequency decomposer is:

[0106] (4)

[0107] Among them, DWT is the discrete wavelet transform, where The SwinTransformer block is used to enhance the deep visual representation capability of the global spectrum. Including high-high, high-low, and low-high frequency components decomposed by DWT, including low-low frequency components;

[0108] S22, extracting global deep visual features in the multi-spectral frequency domain of dual-temporal remote sensing images using a twin architecture based on a step-by-step frequency decomposer and ,Right now:

[0109] = (5)

[0110] S23, align the frequency domain global difference features of the dual-phase high-low multi-spectrum, namely:

[0111] (6)

[0113] in, Select the concatenate function as the frequency domain feature alignment function, These are the frequency domain global difference features of multi-spectrum at four scales from high to low.

[0114] Reference Figures 2 to 4 , S3, by aggregating the local-global space-frequency features of the corresponding scale, construct a multi-scale mixed space-frequency monomer feature, that is:

[0115] (7)

[0116] in, Select the concatenate function as the aggregation function to aggregate space-frequency features, where is the corresponding mixed space-frequency monomer feature of the output.

[0117] S4, designs a random mixed space-frequency monomer fusion module, randomly selects different space-frequency features for fusion, enhances complementarity and calculates contribution weights to activate effective fusion;

[0118] S4 includes a random mixing space-frequency monomer fusion module, the formula is:

[0119] (8)

[0120] in, is a random mixed space-frequency monomer fusion module, where Indicates the frequency of random sampling and the number of spatial monomer experts, is the ith mixed space-frequency monomer, It is a mixed space feature monomer expert library, is the mixed frequency monomer expert library, To output the result;

[0121] Specifically, S4 includes the following steps:

[0122] S41, design random optimization strategy;

[0123] Using the sampling strategy without replacement, n spatial monomer experts are randomly selected from S and n frequency monomer experts are randomly selected from F as auxiliary monomer experts to enhance the mixed spatial-frequency monomer features. ;

[0124] S42, fusion of different spatial / frequency monomer experts to enhance the hybrid spatial / frequency monomer features;

[0125] The input mixed space-frequency monomer feature As the monomer expert to be enhanced, for the randomly selected spatial / frequency monomer expert As an auxiliary monomer expert, where i and m are feature index values;

[0126] S421, align the feature dimensions of the auxiliary single expert and the single expert to be enhanced, namely:

[0127] (9)

[0128] in, is the upsampling convolution function, is the downsampling convolution function, is the concatenate function;

[0129] S422, design a hybrid spatial-frequency attention module, based on Mamba attention, to focus on and enhance the complementarity between different scale and spectral features, namely:

[0130] (10)

[0131] in, is the linear projection layer, It's multi-headed attention. is a multilayer perceptron layer, for:

[0132] (11)

[0133] in, is the linear projection layer, is the convolutional layer, is the SiLU activation function, It is a state space model, which uses vertical mode and horizontal mode to scan data;

[0134] S43, sequentially enhancing the mixed space-frequency monomer features based on step S42 for the randomly selected n space monomer experts and n frequency monomer experts;

[0135] S44, design a hybrid space-frequency monomer expert network, calculate the contribution weight of the fusion one by one through the gated network adaptive calculation, activate the effective features, and eliminate the invalid enhancement features, that is:

[0136] (12)

[0137] in, is the output feature, is the softmax function, which is used to calculate the contribution probability. The optimized mixed space-frequency monomer characteristics in step S43 are: As routing and using the TopK algorithm, the optimal feature is selected as output.

[0138] S5, design a random mixed space-frequency group decision module, randomly select different mixed space-frequency monomers to form a group and perform probabilistic weighting to adaptively select the optimal decision group;

[0139] S5 includes a random mixed space-frequency group decision module, namely:

[0140] (13)

[0141] in, By constructing a mixed space-frequency group To enhance, Represents an enhanced mixed space-frequency monomer library The Features, represents the constructed mixed space-frequency group, and n represents the number of constructed groups. It includes the following steps:

[0142] Specifically, S5 includes the following steps:

[0143] S51, randomly select optimized mixed space-frequency monomer features to form a mixed space-frequency population, that is:

[0144] (14)

[0145] in , p represents the number of monomers randomly drawn from D, q represents the qth population constructed, i is the feature index, and the selected monomers are upsampled or downsampled with The feature dimensions are aligned, and the concatenate function is used to concatenate the features of all selected individuals to construct n groups;

[0146] S52, based on the constructed mixed space-frequency group, makes an optimal decision, namely:

[0147] (15)

[0148] in, For output, The softmax function is used, l represents the lth group, to calculate the contribution probability of each group decision. As a router and using the TopK algorithm, the optimal group decision result is selected as the output;

[0149] S53, using group decision making for optimized hybrid space-frequency monomers The various elements of the group are enhanced.

[0150] S6, builds a lightweight decoder to gradually restore the enhanced multi-scale spatial-frequency features to the resolution of the original image and generate change detection results:

[0151] Specifically, S6 includes the following steps:

[0152] S61, preset lightweight decoder:

[0153] (16)

[0154] in, Represented as an upsampling convolutional layer, For Cat operation, The Mamba attention in step S422 is used to further enhance the feature expression;

[0155] S62, output results;

[0156] (17)

[0157] in, Represented as a convolutional layer, Used to output classification probabilities.

[0158] Reference Figure 5 As shown, this embodiment selects ChangeFormer and P2V-CD network as comparison methods to demonstrate the advantages of visual results in change detection of the present invention.

[0159] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A deep learning remote sensing change detection method based on hybrid space-frequency experts, characterized by: The following steps are involved: S1, selects the twin ConvNeXt network as the spatial domain expert to extract the multi-scale coarse-fine spatial domain local difference features of the dual-temporal remote sensing image; S2, uses twin discrete wavelet transform as the frequency domain expert to extract the high-low frequency domain global difference features of the multi-spectral of the dual-temporal remote sensing image; S3, constructs multi-scale mixed space-frequency monomer features by aggregating local-global space-frequency features of corresponding scales; S4, designs a random mixed space-frequency monomer fusion module, randomly selects different space-frequency features for fusion, enhances complementarity and calculates contribution weights to activate effective fusion; The S4 includes a random mixed space-frequency monomer fusion module, which enhances the characterization capability of mixed space-frequency monomers one by one by fusing different spectrum frequency monomer experts and different scale space monomer experts, namely: (8) in, is a random mixed space-frequency monomer fusion module, where Indicates the frequency of random sampling and the number of spatial monomer experts, is the ith mixed space-frequency monomer, It is a mixed space feature monomer expert library, is the mixed frequency monomer expert library, To output the results, the following steps are included: S41, design random optimization strategy; Adopting the sampling strategy without replacement, n spatial monomer experts are randomly selected from S and n frequency monomer experts are randomly selected from F as auxiliary monomer experts to enhance the mixed spatial-frequency monomer features. ; S42, fusion of different spatial / frequency monomer experts to enhance the hybrid spatial / frequency monomer features; The input mixed space-frequency monomer feature As the monomer expert to be enhanced, for the randomly selected spatial / frequency monomer expert As an auxiliary monomer expert, where i and m are feature index values; S421, align the feature dimensions of the auxiliary single expert and the single expert to be enhanced, namely: (9) in, is the upsampling convolution function, is the downsampling convolution function, is the concatenate function; S422, design a hybrid spatial-frequency attention module, based on Mamba attention, to focus on and enhance the complementarity between different scale and spectral features, namely: (10) in, is the linear projection layer, It's multi-headed attention. is a multilayer perceptron layer, for: (11) in, is the linear projection layer, is the convolutional layer, is the SiLU activation function, It is a state space model, which uses vertical mode and horizontal mode to scan data; S43, sequentially enhancing the mixed space-frequency monomer features based on step S42 for the randomly selected n space monomer experts and n frequency monomer experts; S44, design a hybrid space-frequency monomer expert network, calculate the contribution weight of the fusion one by one through the gated network adaptive calculation, activate the effective features, and eliminate the invalid enhancement features, that is: (12) in, is the output feature, is the softmax function, which is used to calculate the contribution probability. The optimized mixed space-frequency monomer characteristics in step S43 are: As a router, the TopK algorithm is used to select the best features as output; S5, design a random mixed space-frequency group decision module, randomly select different mixed space-frequency monomers to form a group and perform probabilistic weighting to adaptively select the optimal decision group; S6, builds a lightweight decoder to gradually restore the enhanced multi-scale space-frequency features to the resolution of the original image and generate change detection results.

2. The method for remote sensing change detection based on deep learning using hybrid spatial-frequency experts according to claim 1, characterized in that: Said S1 comprises the following steps: S11, extracting the deep spatial features of dual-temporal remote sensing images: ConvNeXt is selected as the spatial domain expert and as the feature encoder, and a weight sharing method is adopted based on the twin architecture to obtain dual-phase remote sensing images. and Multi-scale deep visual features in the spatial domain and ,Right now: (1) S12, aligns the dual-temporal multi-scale coarse-fine spatial domain local difference features, namely: (2) in, Select the concatenate function for multi-dimensional spatiotemporal alignment. There are four multi-scale spatial local difference features from coarse to fine.

3. The deep learning remote sensing change detection method based on hybrid space-frequency experts according to claim 1 is characterized by: The S2 comprises the following steps: S21, using the step-by-step decoupling method to select discrete wavelet transform to extract the multi-spectral frequency domain features of single-phase remote sensing images, namely: (3) in, The detailed iterative process of the step-by-step frequency decomposer is: (4) Among them, DWT is the discrete wavelet transform, where The SwinTransformer block is used to enhance the deep visual representation capability of the global spectrum. Including high-high, high-low, and low-high frequency components decomposed by DWT, including low-low frequency components; S22, using a twin architecture based on a step-by-step frequency decomposer to extract global deep visual features in the multi-spectral frequency domain of dual-temporal remote sensing images and ,Right now: = (5) S23, align the frequency domain global difference features of the dual-phase high-low multi-spectrum, namely: (6) in, Select the concatenate function as the frequency domain feature alignment function, These are the frequency domain global difference features of multi-spectrum at four scales from high to low.

4. The method for remote sensing change detection based on deep learning using hybrid space-frequency experts according to claim 1, characterized in that: The S3 is: The corresponding multi-scale spatial domain local difference features and multi-spectral frequency domain global difference features are fused through the aggregation process and used as mixed space-frequency monomer features, namely: (7) in, Select the concatenate function as the aggregation function to aggregate space-frequency features, where is the corresponding mixed space-frequency monomer feature of the output.

5. The method for remote sensing change detection based on deep learning using hybrid space-frequency experts according to claim 1, characterized in that: The design random mixed space-frequency group decision module in S5 is: (13) in, By constructing a mixed space-frequency group To enhance, Represents an enhanced mixed space-frequency monomer library The Features, represents the constructed mixed space-frequency group, n represents the number of constructed groups, and includes the following steps: S51, randomly select optimized mixed space-frequency monomer features to form a mixed space-frequency population, that is: (14) in , p represents the number of monomers randomly drawn from D, q represents the qth population constructed, i is the feature index, and the selected monomers are upsampled or downsampled with The feature dimensions of the selected individuals are aligned, and the concatenate function is used to concatenate the features of all the selected individuals to construct n groups; S52, based on the constructed mixed space-frequency group, makes an optimal decision, namely: (15) in, For output, The softmax function is used, l represents the lth group, to calculate the contribution probability of each group decision. As a router and using the TopK algorithm, the optimal group decision result is selected as the output; S53, using group decision making for optimized hybrid space-frequency monomers The various elements of the group are enhanced.

6. The method for remote sensing change detection based on deep learning using hybrid space-frequency experts according to claim 1, characterized in that: The S6 comprises the following steps: S61, preset lightweight decoder: (16) in, Represented as an upsampling convolutional layer, For Cat operation, The Mamba attention in step S422 is used to further enhance the feature expression; S62, output results; (17) in, Represented as a convolutional layer, Used to output classification probabilities.

Images