A Remote Sensing Change Detection Method for Urban Residential Spaces Based on Deep Learning

Abstract

The present invention discloses a remote sensing change detection method for urban residential spaces based on deep learning, which relates to the technical field of remote sensing image processing. First, a hierarchical densely connected network module is used to respectively extract deep features from remote sensing images of urban residential spaces at two different time phases, and shallow, middle, and high-level features are output. Then, a Transformer network module is used to perform self-attention feature mining on the shallow, middle, and high-level features to highlight the significant features of the urban residential spaces in the time phase. Next, a hierarchical feature cross-attention mechanism is used to fuse the shallow, middle, and high-level feature information. Finally, a multi-layer perceptron is used to identify and detect the changed and unchanged urban residential spaces in the remote sensing images of the two time phases, and a change detection map is output. The present invention can enhance the ability of the network model to extract and detect the temporal change features of urban residential spaces, improve the accuracy of the detection results, and provide a refined empirical basis for application fields such as urban planning, optimization of urban functional spaces, and evolution of urban residential spaces.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and more specifically, to a method for remote sensing change detection of urban residential space based on deep learning. Background Art

[0002] The urban residential space change detection technology is a method for identifying and quantifying the changes in residential land in urban areas by using remote sensing data. The advantage of this technology lies in its ability to provide large-scale, high-timeliness, and objective residential space change information, and it is widely used in fields such as urban planning, land management, and environmental monitoring. Currently, the existing spatial change detection technologies are often applied to the field of urban space expansion research, aiming to explore the area, scope, and speed of the outward expansion of the urban built-up area during the process of urbanization. The common identification methods for urban space mainly include the method of comprehensively using remote sensing technology and geographic information system, and the spatial detection method based on the cellular automata model.

[0003] However, the current detection of urban residential space evolution is mostly included in the research of urban functional areas. In terms of detection technology, most of them focus on the big data-driven model based on GIS. However, its data input is time-consuming, the data quality is uneven, and the analysis results highly depend on the quality of the original data, making the interpretation of the analysis results tend to be complicated. The existing urban residential space change detection technologies often rely on prior knowledge or parameter adjustment to adapt to specific scenarios, and it is difficult to generalize to new environments. Moreover, their performance is limited by the feature expression ability and may not be able to fully express the complex changes in urban residential space. The current remote sensing detection technologies are often used to detect the spatial changes in urban built-up areas, and the detection scale is relatively macroscopic, without focusing on the change trend of a specific type of space in the city, such as residential space. Therefore, from a microscopic perspective, the accuracy and reliability of the existing remote sensing detection technologies need to be improved. Summary of the Invention

[0004] In order to overcome the defect of low accuracy in detecting the temporal change characteristics of urban residential space, the present invention provides a method for remote sensing change detection of urban residential space based on deep learning.

[0005] To solve the above technical problems, the technical solution of the present invention is as follows:

[0006] The present invention proposes a method for remote sensing change detection of urban residential space based on deep learning, including:

[0007] S1: Obtain a set of remote sensing image pairs of the residential space in the target city. Each pair of remote sensing images in the set of remote sensing image pairs includes a first-phase remote sensing image and a second-phase remote sensing image. Obtain the corresponding real change image according to the remote sensing image pair, and construct a remote sensing change detection model for the urban residential space. The change detection model includes a hierarchical dense connection network module, a Transformer network module, a cross-attention module, and a multi-layer perceptron module;

[0008] S2: Input the first-phase remote sensing image and the second-phase remote sensing image into the hierarchical dense connection network module respectively to obtain a first-phase multi-scale feature image and a second-phase multi-scale feature image;

[0009] S3: Input the first-phase multi-scale feature image and the second-phase multi-scale feature image into the Transformer network module for feature mining to obtain a first-phase multi-scale significant feature image and a second-phase multi-scale significant feature image;

[0010] S4: Input the first-phase multi-scale significant feature image and the second-phase multi-scale significant feature image into the cross-attention module for fusion to obtain a fusion image, and input the fusion image into the multi-layer perceptron module to obtain a predicted change image;

[0011] S5: Construct a total loss function according to the real change image and the predicted change image, and train the remote sensing change detection model for the urban residential space to obtain a trained remote sensing change detection model for the urban residential space;

[0012] S6: Obtain remote sensing images of different phases of the residential space in the target city, and input them into the trained remote sensing change detection model for the urban residential space to obtain a change detection image.

[0013] Preferably, in S1, the obtaining the corresponding real change image according to the remote sensing image pair includes:

[0014] For the first-phase remote sensing image and the second-phase remote sensing image in the remote sensing image pair, if there is a difference in ground object coverage in the same pixel of the first-phase remote sensing image and the second-phase remote sensing image, mark the corresponding pixel as 1, otherwise mark it as 0;

[0015] All pixels form the real change image.

[0016] Preferably, in S1, the hierarchical dense connection network module includes a first BN-ReLU-Conv2D sub-module, a first Concat layer, a second BN-ReLU-Conv2D sub-module, a second Concat layer, a third BN-ReLU-Conv2D sub-module, a third Concat layer, a fourth BN-ReLU-Conv2D sub-module, and a fourth Concat layer connected in sequence;

[0017] The input end of the first BN-ReLU-Conv2D sub-module is also connected to the input ends of the first Concat layer, the second Concat layer, the third Concat layer, and the fourth Concat layer;

[0018] The output end of the first BN-ReLU-Conv2D sub-module is also connected to the input ends of the second Concat layer, the third Concat layer, and the fourth Concat layer;

[0019] The output end of the second BN-ReLU-Conv2D sub-module is also connected to the input ends of the third Concat layer and the fourth Concat layer;

[0020] The output end of the third BN-ReLU-Conv2D sub-module is also connected to the input end of the fourth Concat layer;

[0021] The output end of the fourth Concat layer is connected to the input end of the Transformer network module.

[0022] Preferably, the BN-ReLU-Conv2D sub-module includes a batch normalization layer, a ReLu activation function layer, and a two-dimensional convolutional layer connected in sequence:

[0023] Preferably, in the S1, the Transformer network module includes L feature supervision sub-modules connected in sequence;

[0024] Each of the feature supervision sub-modules includes an Embedded Patches layer, a first normalization layer, a multi-head self-attention layer, a first splicing layer, a second normalization layer, a multi-layer perceptron, and a second splicing layer connected in sequence;

[0025] The output end of the Embedded Patches layer is also connected to the input end of the first splicing layer, and the output end of the first splicing layer is also connected to the input end of the second splicing layer;

[0026] The input end of the Embedded Patches layer of the first feature supervision sub-module is connected to the output end of the hierarchical densely connected network module.

[0027] Preferably, in the S1, the cross-attention module includes a first 1×1 convolutional layer, a second 1×1 convolutional layer, a third 1×1 convolutional layer, an Affinity layer, a softmax layer, and an Aggregation layer;

[0028] The first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are arranged in parallel, and the input ends of the first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are all connected to the output end of the Transformer network module;

[0029] The output end of the first 1×1 convolutional layer is connected to the first input end of the Aggregation layer, the output end of the second 1×1 convolutional layer is connected to the first input end of the Affinity layer, the output end of the third 1×1 convolutional layer is connected to the second input end of the Affinity layer, the output end of the Affinity layer is connected to the input end of the softmax layer, and the output end of the softmax layer is connected to the second input end of the Aggregation layer;

[0030] Preferably, the method for determining the total loss function includes:

[0031]

[0032] Among them, represents the total loss function, N represents the size of the set of remote sensing image pairs of the target urban residential space, and Y b represents the true change image of the b-th pair of remote sensing image pairs, represents the predicted change image of the b-th pair of remote sensing image pairs.

[0033] The present invention also proposes a deep learning-based urban residential space remote sensing change detection device, including:

[0034] A data acquisition module for acquiring a set of remote sensing image pairs of the target urban residential space, each set of remote sensing image pairs in the set of remote sensing image pairs includes a first-phase remote sensing image and a second-phase remote sensing image, obtaining a corresponding true change image according to the remote sensing image pairs, and constructing an urban residential space remote sensing change detection model, the change detection model including a hierarchical dense connection network module, a Transformer network module, a cross-attention module, and a multi-layer perceptron module;

[0035] A multi-scale feature module for respectively inputting the first-phase remote sensing image and the second-phase remote sensing image into the hierarchical dense connection network module to obtain a first-phase multi-scale feature image and a second-phase multi-scale feature image;

[0036] A multi-scale significant feature module for inputting the first-phase multi-scale feature image and the second-phase multi-scale feature image into the Transformer network module for feature mining to obtain a first-phase multi-scale significant feature image and a second-phase multi-scale significant feature image;

[0037] A fusion module, which is used to input the multi-scale salient feature images of the first phase and the multi-scale salient feature images of the second phase into a cross-attention module for fusion to obtain a fused image, and input the fused image into a multi-layer perceptron module to obtain a predicted change image;

[0038] A model training module, which is used to construct a total loss function according to the real change image and the predicted change image, train the remote sensing change detection model for urban residential space, and obtain a trained remote sensing change detection model for urban residential space;

[0039] A change detection image acquisition module, which is used to acquire remote sensing images of different phases of the target urban residential space, input them into the trained remote sensing change detection model for urban residential space, and obtain a change detection image.

[0040] The present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the above detection method are implemented.

[0041] The present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above detection method are implemented.

[0042] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0043] The present invention provides a remote sensing change detection method for urban residential space based on deep learning. First, a hierarchical dense connection network module is used to perform deep feature extraction on remote sensing images of urban residential space in two consecutive phases, and shallow, middle, and high-level features are output; then, a Transformer network module is used to perform self-attention feature mining on the shallow, middle, and high-level features to highlight the salient features of the urban residential space in the phase. Then, a hierarchical feature cross-attention mechanism is used to fuse the shallow, middle, and high-level feature information. Finally, a multi-layer perceptron is used to identify and detect the changed and unchanged urban residential spaces in the remote sensing images of the two phases, and a change detection map is output. The present invention can enhance the ability of the network model to extract and detect the temporal change features of urban residential space, improve the accuracy of the detection results, and provide a refined empirical basis for application fields such as urban planning, urban functional space optimization, and urban residential space evolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic flowchart of the remote sensing change detection method described in Embodiment 1;

[0045] Figure 2 It is an overall block diagram of the remote sensing change detection method described in Embodiment 2;

[0046] Figure 3The figure shows the change detection result of the remote sensing image pair of the area near Dashadi, Huangpu District, Guangzhou in 2015 and 2020 by the remote sensing change detection method described in Embodiment 2;

[0047] Figure 4 The figure shows the schematic diagram of the computer device described in Embodiment 2;

[0048] Figure 5 The figure shows the structural schematic diagram of the remote sensing change detection device described in Embodiment 3. Detailed implementation manners

[0049] The accompanying drawings are only for illustrative purposes and should not be construed as limitations on this patent;

[0050] To better illustrate this embodiment, some components in the accompanying drawings are omitted, enlarged or reduced, and do not represent the dimensions of the actual product;

[0051] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0052] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0053] Embodiment 1

[0054] This embodiment provides a remote sensing change detection method for urban residential space based on deep learning, as Figure 1 shown, including:

[0055] S1: Obtain a set of remote sensing image pairs of the target urban residential space. Each remote sensing image pair in the set of remote sensing image pairs includes a first-phase remote sensing image and a second-phase remote sensing image. Obtain the corresponding true change image according to the remote sensing image pair, and construct a remote sensing change detection model for urban residential space. The change detection model includes a hierarchical dense connection network module, a Transformer network module, a cross-attention module, and a multi-layer perceptron module;

[0056] S2: Input the first-phase remote sensing image and the second-phase remote sensing image into the hierarchical dense connection network module respectively to obtain a first-phase multi-scale feature image and a second-phase multi-scale feature image;

[0057] S3: Input the first-phase multi-scale feature image and the second-phase multi-scale feature image into the Transformer network module for feature mining to obtain a first-phase multi-scale significant feature image and a second-phase multi-scale significant feature image;

[0058] S4: Input the first-phase multi-scale significant feature image and the second-phase multi-scale significant feature image into the cross-attention module for fusion to obtain a fusion image, and input the fusion image into the multi-layer perceptron module to obtain a predicted change image;

[0059] S5: Construct a total loss function based on the real change image and the predicted change image, train the remote sensing change detection model for urban residential space, and obtain a trained remote sensing change detection model for urban residential space.

[0060] S6: Obtain remote sensing images of different time phases of the target urban residential space, and input them into the trained remote sensing change detection model for urban residential space to obtain a change detection image.

[0061] In the specific implementation process, first obtain a pair of remote sensing images of the target urban residential space, construct a remote sensing change detection model for urban residential space, input the pair of remote sensing images into the hierarchical dense connection network module to obtain the first-phase multi-scale feature image and the second-phase multi-scale feature image, then input them into the Transformer network module for feature mining to obtain the first-phase multi-scale significant feature image and the second-phase multi-scale significant feature image. After that, input them into the cross-attention module for fusion to obtain a fused image, then input the fused image into the multi-layer perceptron module to obtain a predicted change image; finally, construct a total loss function based on the real change image and the predicted change image, train and optimize the remote sensing change detection model for urban residential space, obtain remote sensing images of different time phases of the target urban residential space, and input them into the trained remote sensing change detection model for urban residential space to obtain a change detection image.

[0062] Embodiment 2

[0063] The present invention proposes a method for remote sensing change detection of urban residential space based on deep learning, including:

[0064] S1: Obtain a set of pairs of remote sensing images of the target urban residential space, each pair of remote sensing images in the set of pairs of remote sensing images includes a first-phase remote sensing image and a second-phase remote sensing image, obtain a corresponding real change image according to the pair of remote sensing images, and construct a remote sensing change detection model for urban residential space. The change detection model includes a hierarchical dense connection network module, a Transformer network module, a cross-attention module, and a multi-layer perceptron module.

[0065] The obtaining of the corresponding real change image according to the pair of remote sensing images includes:

[0066] For the first-phase remote sensing image and the second-phase remote sensing image in the pair of remote sensing images, if there is a difference in ground object coverage at the same pixel point of the first-phase remote sensing image and the second-phase remote sensing image, mark the corresponding pixel point as 1, otherwise mark it as 0.

[0067] All pixel points form a real change image.

[0068] Such as Figure 2As shown, it is the overall block diagram of the remote sensing change detection method proposed in this embodiment.

[0069] The hierarchical densely connected network module includes a first BN-ReLU-Conv2D sub-module, a first Concat layer, a second BN-ReLU-Conv2D sub-module, a second Concat layer, a third BN-ReLU-Conv2D sub-module, a third Concat layer, a fourth BN-ReLU-Conv2D sub-module, and a fourth Concat layer connected in sequence;

[0070] The input end of the first BN-ReLU-Conv2D sub-module is also connected to the input ends of the first Concat layer, the second Concat layer, the third Concat layer, and the fourth Concat layer;

[0071] The output end of the first BN-ReLU-Conv2D sub-module is also connected to the input ends of the second Concat layer, the third Concat layer, and the fourth Concat layer;

[0072] The output end of the second BN-ReLU-Conv2D sub-module is also connected to the input ends of the third Concat layer and the fourth Concat layer;

[0073] The output end of the third BN-ReLU-Conv2D sub-module is also connected to the input end of the fourth Concat layer;

[0074] The output end of the fourth Concat layer is connected to the input end of the Transformer network module.

[0075] The BN-ReLU-Conv2D sub-module includes a batch normalization layer, a ReLu activation function layer, and a two-dimensional convolutional layer connected in sequence:

[0076] The Transformer network module includes L feature supervision sub-modules connected in sequence;

[0077] Each of the feature supervision sub-modules includes an Embedded Patches layer, a first normalization layer, a multi-head self-attention layer, a first splicing layer, a second normalization layer, a multi-layer perceptron, and a second splicing layer connected in sequence;

[0078] The output end of the Embedded Patches layer is also connected to the input end of the first splicing layer, and the output end of the first splicing layer is also connected to the input end of the second splicing layer;

[0079] The input end of the Embedded Patches layer of the first feature supervision sub-module is connected to the output end of the hierarchical dense connection network module.

[0080] The multi-head self-attention layer captures complex relationships between sequences by processing multiple attention heads in parallel. It allows the model to capture information in different representation subspaces simultaneously. Each head performs its own attention calculation, and then the outputs of these heads are combined to provide a more comprehensive image representation. In practical applications, the multi-head self-attention is replicated multiple times. Each head calculates the attention in a different representation subspace, and then the outputs of these heads are concatenated or averaged to form the final output. The layer normalization layer is used to normalize the feature representations within each layer. This normalization technique helps reduce the internal covariate shift, thereby enhancing the stability of the model learning process.

[0081] The cross-attention module includes a first 1×1 convolutional layer, a second 1×1 convolutional layer, a third 1×1 convolutional layer, an Affinity layer, a softmax layer, and an Aggregation layer; the softmax function ensures that the sum of the output attention weights is 1, so that an effective weighted sum can be obtained.

[0082] The first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are arranged in parallel. The input ends of the first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are all connected to the output end of the Transformer network module;

[0083] The output end of the first 1×1 convolutional layer is connected to the first input end of the Aggregation layer. The output end of the second 1×1 convolutional layer is connected to the first input end of the Affinity layer. The output end of the third 1×1 convolutional layer is connected to the second input end of the Affinity layer. The output end of the Affinity layer is connected to the input end of the softmax layer. The output end of the softmax layer is connected to the second input end of the Aggregation layer;

[0084] In image processing, cross-attention can be used to enhance the model's perception of the relationships between different regions in an image. For example, when identifying an object in an image, the model not only focuses on the features of the object itself but also on the relationships between the object and the background or other objects. This embodiment adopts a shallow, middle, and high-level feature cross-attention mechanism to fully fuse and interact the feature information of each level.

[0085] z l ′ = MSA(LN(EP(z l-1 )))+z l-1 l = 2,…,L

[0086] z l= MLP(LN(z l ′)) + z l ′ for l = 2, ..., L

[0087] where z l-1 represents the output of the second concatenation layer of the (l - 1)-th feature supervision sub-module, and z l ′ represents the output of the first concatenation layer of the l-th feature supervision sub-module. MSA represents the multi-head self-attention processing function, LN represents the layer normalization processing function, MLP represents the multi-layer perceptron processing function, and EP represents the Embedded Patches processing operation.

[0088] S2: Input the first-phase remote sensing image and the second-phase remote sensing image into the hierarchical dense connection network module respectively to obtain the first-phase multi-scale feature image and the second-phase multi-scale feature image;

[0089] S3: Input the first-phase multi-scale feature image and the second-phase multi-scale feature image into the Transformer network module for feature mining to obtain the first-phase multi-scale significant feature image and the second-phase multi-scale significant feature image;

[0090] S4: Input the first-phase multi-scale significant feature image and the second-phase multi-scale significant feature image into the cross-attention module for fusion to obtain a fusion image, and input the fusion image into the multi-layer perceptron module to obtain a predicted change image;

[0091] S5: Construct a total loss function based on the true change image and the predicted change image, and train the remote sensing change detection model for urban residential space to obtain a trained remote sensing change detection model for urban residential space;

[0092] The method for determining the total loss function includes:

[0093]

[0094] where represents the total loss function, N represents the size of the set of target urban residential space remote sensing image pairs, Y b represents the true change image of the b-th pair of remote sensing image pairs, represents the predicted change image of the b-th pair of remote sensing image pairs.

[0095] S6: Obtain remote sensing images of different phases of the target urban residential space, input them into the trained remote sensing change detection model for urban residential space, and obtain a change detection image.

[0096] The detection method proposed in this embodiment is used to verify the remote sensing image pairs of the area near Dashadi, Huangpu District, Guangzhou in 2015 and 2020. The residential space in this area mainly includes commercial housing residential communities along urban main roads and subway lines, self-built houses in the middle of residential plots, urban villages, etc. As Figure 3 shown, it is the result map of urban residential space change detection in this area. It can be seen from the results that in the residential space, due to the construction needs of the community commercial comprehensive carrier, the commercial housing community has changed greatly near the urban road intersections; the self-built houses and urban villages have changed greatly in the adjacent areas to other urban functional spaces such as surrounding residential communities, commercial complexes, and factory areas. In addition, this embodiment evaluates the effectiveness of the technical method quantitatively. Table 1 gives the performance evaluation indicators of this embodiment in the urban residential space change detection task, including F1 score (F1), intersection over union (IoU), precision (Pre), and recall (Rec). The F1 score is the harmonic mean of precision and recall. It attempts to achieve a balance between precision and recall, and its value range is [0,1]. In many fields, an F1 score of 0.7 or higher is generally considered to have good performance. IoU measures the degree of overlap between the predicted segmentation area and the true segmentation area, also known as the Jaccard index. In the image segmentation task, IoU usually needs to reach 0.5 or higher to be considered reasonable performance. Precision and recall are generally considered to be in opposition to each other, and this phenomenon is called the Precision-Recall Trade-off in statistics and machine learning. Precision focuses on the proportion of actual positive classes among the predicted positive classes, that is, reducing false positives. Generally speaking, a precision of 0.8 or above is considered good. Recall focuses on the proportion of all actual positive classes that are correctly predicted as positive classes, that is, reducing false negatives. Similarly, a recall of 0.8 or above is usually considered good. It can be seen from Table 1 that the F1 score (F1), intersection over union (IoU), precision (Pre), and recall (Rec) of the detection method proposed in this embodiment far exceed the general measurement values, and it has very high detection accuracy.

[0097] Table 1 Verification parameters for change detection results

[0098] Verification Index F1 IoU Pre Rec CFCANet 94.03％ 89.63％ 95.25％ 93.82％

[0099] This embodiment also proposes a computer device, as Figure 4 shown, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above detection method.

[0100] This embodiment also proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, it implements the steps of the above detection method.

[0101] Example 3

[0102] This embodiment provides a remote sensing change detection device for urban residential space based on deep learning, as Figure 5 shown, including:

[0103] A data acquisition module, configured to acquire a set of remote sensing image pairs of the target urban residential space. Each remote sensing image pair in the set of remote sensing image pairs includes a first-phase remote sensing image and a second-phase remote sensing image. According to the remote sensing image pairs, a corresponding true change image is acquired, and a remote sensing change detection model for urban residential space is constructed. The change detection model includes a hierarchical densely connected network module, a Transformer network module, a cross-attention module, and a multi-layer perceptron module;

[0104] A multi-scale feature module, configured to input the first-phase remote sensing image and the second-phase remote sensing image into the hierarchical densely connected network module respectively, to obtain a first-phase multi-scale feature image and a second-phase multi-scale feature image;

[0105] A multi-scale significant feature module, configured to input the first-phase multi-scale feature image and the second-phase multi-scale feature image into the Transformer network module for feature mining, to obtain a first-phase multi-scale significant feature image and a second-phase multi-scale significant feature image;

[0106] A fusion module, configured to input the first-phase multi-scale significant feature image and the second-phase multi-scale significant feature image into the cross-attention module for fusion, to obtain a fusion image, and input the fusion image into the multi-layer perceptron module, to obtain a predicted change image;

[0107] A model training module, configured to construct a total loss function according to the true change image and the predicted change image, and train the remote sensing change detection model for urban residential space to obtain a trained remote sensing change detection model for urban residential space;

[0108] A change detection image acquisition module, configured to acquire remote sensing images of different phases of the target urban residential space, and input them into the trained remote sensing change detection model for urban residential space to obtain a change detection image.

[0109] The same or similar reference numerals correspond to the same or similar components;

[0110] The terms describing the positional relationship in the drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0111] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A method for detecting changes in urban residential space using remote sensing based on deep learning, characterized in that: include: S1: Obtain a set of remote sensing image pairs of the target urban residential space, wherein each set of remote sensing image pairs in the remote sensing image pair set includes a first phase remote sensing image and a second phase remote sensing image, obtain corresponding real change images according to the remote sensing image pairs, and construct a remote sensing change detection model for urban residential space, wherein the change detection model includes a hierarchical densely connected network module, a Transformer network module, a cross attention module, and a multi-layer perceptron module; S2: inputting the first time phase remote sensing image and the second time phase remote sensing image into the hierarchical dense connection network module respectively to obtain the first time phase multi-scale feature image and the second time phase multi-scale feature image; S3: Inputting the first phase multi-scale feature image and the second phase multi-scale feature image into the Transformer network module for feature mining to obtain the first phase multi-scale significant feature image and the second phase multi-scale significant feature image; S4: inputting the first phase multi-scale salient feature image and the second phase multi-scale salient feature image into the cross attention module for fusion to obtain a fused image, and inputting the fused image into the multi-layer perceptron module to obtain a predicted change image; S5: constructing a total loss function according to the real change image and the predicted change image, training the urban residential space remote sensing change detection model, and obtaining a trained urban residential space remote sensing change detection model; S6: Obtain remote sensing images of the target city’s residential space at different times, and input them into the trained urban residential space remote sensing change detection model to obtain change detection images; In S1, the hierarchical densely connected network module includes a first BN-ReLU-Conv2D submodule, a first Concat layer, a second BN-ReLU-Conv2D submodule, a second Concat layer, a third BN-ReLU-Conv2D submodule, a third Concat layer, a fourth BN-ReLU-Conv2D submodule and a fourth Concat layer connected in sequence; The input end of the first BN-ReLU-Conv2D submodule is also connected to the input end of the first Concat layer, the input end of the second Concat layer, the input end of the third Concat layer, and the input end of the fourth Concat layer; The output end of the first BN-ReLU-Conv2D submodule is also connected to the input end of the second Concat layer, the input end of the third Concat layer, and the input end of the fourth Concat layer; The output end of the second BN-ReLU-Conv2D submodule is also connected to the input end of the third Concat layer and the input end of the fourth Concat layer; The output of the third BN-ReLU-Conv2D submodule is also connected to the input of the fourth Concat layer; The output of the fourth Concat layer is connected to the input of the Transformer network module.

2. The method for detecting changes in urban residential space based on remote sensing based on deep learning according to claim 1, characterized in that: In S1, obtaining the corresponding real change image according to the remote sensing image pair includes: For the first phase remote sensing image and the second phase remote sensing image in the remote sensing image pair, if there is a difference in ground object coverage in the same pixel point between the first phase remote sensing image and the second phase remote sensing image, the corresponding pixel point is marked as 1, otherwise it is marked as 0; All pixels constitute the real change image.

3. The method for detecting changes in urban residential space based on remote sensing based on deep learning according to claim 1, characterized in that: The BN-ReLU-Conv2D submodule includes a batch normalization layer, a ReLu activation function layer and a two-dimensional convolution layer connected in sequence.

4. The method for detecting changes in urban residential space based on remote sensing based on deep learning according to claim 2 is characterized in that: In S1, the Transformer network module includes L feature supervision submodules connected in sequence; Each of the feature supervision submodules includes an Embedded Patches layer, a first normalization layer, a multi-head self-attention layer, a first splicing layer, a second normalization layer, a multi-layer perceptron and a second splicing layer connected in sequence; The output end of the Embedded Patches layer is also connected to the input end of the first splicing layer, and the output end of the first splicing layer is also connected to the input end of the second splicing layer; The input of the Embedded Patches layer of the first feature supervision submodule is connected to the output of the hierarchical densely connected network module.

5. The method for detecting changes in urban residential space using remote sensing based on deep learning according to claim 1, characterized in that: In S1, the cross attention module includes a first 1×1 convolutional layer, a second 1×1 convolutional layer, a third 1×1 convolutional layer, an Affinity layer, a softmax layer and an Aggregation layer; The first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are arranged in parallel, and the input ends of the first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are all connected to the output end of the Transformer network module; The output of the first 1×1 convolutional layer is connected to the first input of the Aggregation layer, the output of the second 1×1 convolutional layer is connected to the first input of the Affinity layer, the output of the third 1×1 convolutional layer is connected to the second input of the Affinity layer, the output of the Affinity layer is connected to the input of the softmax layer, and the output of the softmax layer is connected to the second input of the Aggregation layer.

6. The method for detecting changes in urban residential space based on remote sensing based on deep learning according to claim 5 is characterized in that: The method for determining the total loss function includes: in, represents the total loss function, N represents the size of the target city residential space remote sensing image set, Y b represents the real change image of the bth pair of remote sensing images, Represents the predicted change image of the bth pair of remote sensing images.

7. A device for detecting changes in urban residential space based on remote sensing based on deep learning, characterized in that: include: A data acquisition module is used to acquire a set of remote sensing image pairs of the target urban residential space, wherein each set of remote sensing image pairs in the remote sensing image pair set includes a remote sensing image of a first time phase and a remote sensing image of a second time phase, and corresponding real change images are acquired according to the remote sensing image pairs to construct a remote sensing change detection model for urban residential space, wherein the change detection model includes a hierarchical densely connected network module, a Transformer network module, a cross attention module, and a multi-layer perceptron module; A multi-scale feature module, used to input the first phase remote sensing image and the second phase remote sensing image into the hierarchical dense connection network module respectively, to obtain the first phase multi-scale feature image and the second phase multi-scale feature image; A multi-scale significant feature module is used to input the first phase multi-scale feature image and the second phase multi-scale feature image into the Transformer network module for feature mining to obtain the first phase multi-scale significant feature image and the second phase multi-scale significant feature image; A fusion module is used to input the first phase multi-scale significant feature image and the second phase multi-scale significant feature image into the cross attention module for fusion to obtain a fused image, and input the fused image into the multi-layer perceptron module to obtain a predicted change image; A model training module is used to construct a total loss function according to the real change image and the predicted change image, train the urban residential space remote sensing change detection model, and obtain a trained urban residential space remote sensing change detection model; The change detection image acquisition module is used to obtain remote sensing images of the target urban residential space at different times and input them into the trained urban residential space remote sensing change detection model to obtain change detection images; The hierarchical densely connected network module includes a first BN-ReLU-Conv2D submodule, a first Concat layer, a second BN-ReLU-Conv2D submodule, a second Concat layer, a third BN-ReLU-Conv2D submodule, a third Concat layer, a fourth BN-ReLU-Conv2D submodule and a fourth Concat layer connected in sequence; The input end of the first BN-ReLU-Conv2D submodule is also connected to the input end of the first Concat layer, the input end of the second Concat layer, the input end of the third Concat layer, and the input end of the fourth Concat layer; The output end of the first BN-ReLU-Conv2D submodule is also connected to the input end of the second Concat layer, the input end of the third Concat layer, and the input end of the fourth Concat layer; The output end of the second BN-ReLU-Conv2D submodule is also connected to the input end of the third Concat layer and the input end of the fourth Concat layer; The output of the third BN-ReLU-Conv2D submodule is also connected to the input of the fourth Concat layer; The output of the fourth Concat layer is connected to the input of the Transformer network module.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the detection method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the detection method according to any one of claims 1 to 6 are implemented.

Images