Remote sensing image change detection method

Abstract

The invention discloses a remote sensing image change detection method, and belongs to the technical field of remote sensing image processing, and the method comprises the following steps: obtaining a to-be-detected dual-time-phase remote sensing image which comprises a front-time-phase remote sensing image and a rear-time-phase remote sensing image; inputting the dual-time-phase remote sensing image into a pre-trained improved change detection network model to obtain a change region and a change type of the dual-time-phase remote sensing image; wherein a Shuffled-Sparse Attention module and a CFFEM module are introduced into the improved type change detection network model, and the improved type change detection network model is used for detecting the change of the network. According to the method, finer boundary information can be extracted, the omission ratio and the error detection ratio are reduced, and the model generalization is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and more particularly to a remote sensing image change detection method. Background Art

[0002] Remote sensing image change detection is an important research field in geographic information systems (GIS) and remote sensing (RS). It detects changes in surface cover by analyzing remote sensing images of different time or spatial resolutions, which provides important decision support for environmental monitoring, urban planning, land resource utilization, disaster warning, etc.

[0003] Currently, the common deep learning-based change detection methods for remote sensing images mainly focus on extracting change detection information through twin neural networks and attention mechanism models. Although they have achieved outstanding results in a variety of data sets, they still have the following problems: 1) weak extraction of boundary information; 2) missed detections and false detections; 3) inconsistent extraction accuracy of the model on different ground object change detection data sets, resulting in low generalization.

[0004] Therefore, how to provide a remote sensing image change detection method that can extract more detailed boundary information, reduce missed detection rate and false detection rate, and improve model generalization is a problem that technicians in this field urgently need to solve. Summary of the invention

[0005] In view of this, an object of the present invention is to provide a remote sensing image change detection method.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A remote sensing image change detection method comprises the following steps:

[0008] Acquire the dual-temporal remote sensing image to be detected: wherein the dual-temporal remote sensing image to be detected includes the front-phase remote sensing image and the back-phase remote sensing image;

[0009] The dual-temporal remote sensing image is input into a pre-trained improved change detection network model to obtain the change area and change type of the dual-temporal remote sensing image; wherein the improved change detection network model introduces a Shuffled-Sparse Attention module and a CFFEM module.

[0010] Preferably, the improved change detection network model includes an encoder;

[0011] The encoder includes a first downsampling module, a first Shuffled-Sparse Attention module, a second downsampling module, a second Shuffled-Sparse Attention module, a third downsampling module, a third Shuffled-Sparse Attention module, a fourth downsampling module, and a fourth Shuffled-Sparse Attention module, which are sequentially connected;

[0012] The encoder further includes a fifth downsampling module, a fifth Shuffled-SparseAttention module, a sixth downsampling module, a sixth Shuffled-Sparse Attention module, a seventh downsampling module, a seventh Shuffled-Sparse Attention module, an eighth downsampling module, and an eighth Shuffled-SparseAttention module connected in sequence;

[0013] The input end of the first down-sampling module is connected to the remote sensing image of the previous phase;

[0014] The input end of the fifth down-sampling module is connected to the later phase remote sensing image;

[0015] The encoder also includes a first CFFEM module, a second CFFEM module, a third CFFEM module and a fourth CFFEM module;

[0016] The output end of the first Shuffled-Sparse Attention module and the output end of the fifth Shuffled-Sparse Attention module are both connected to the input end of the first CFFEM module;

[0017] The output end of the second Shuffled-Sparse Attention module and the output end of the sixth Shuffled-Sparse Attention module are both connected to the input end of the second CFFEM module;

[0018] The output end of the third Shuffled-Sparse Attention module and the output end of the seventh Shuffled-Sparse Attention module are both connected to the input end of the third CFFEM module;

[0019] The output end of the fourth Shuffled-Sparse Attention module and the output end of the eighth Shuffled-Sparse Attention module are both connected to the input end of the fourth CFFEM module.

[0020] Preferably, each Shuffled-Sparse Attention module includes a first Norm layer, a shuffledAttention Layer, a first addition unit, a second Norm layer, an MLP layer, and a second addition unit;

[0021] The input end of the first Norm layer is the input end of the Shuffled-Sparse Attention module;

[0022] The output end of the second adding unit is the output end of the Shuffled-Sparse Attention module;

[0023] An input end of the first Norm layer is connected to an input end of the first adding unit;

[0024] The output end of the first Norm layer is connected to the input end of the shuffled Attention Layer;

[0025] An output end of the shuffled Attention Layer is connected to an input end of the first adding unit;

[0026] The output end of the first adding unit is connected to the input end of the second Norm layer and the second adding unit respectively;

[0027] The output end of the second Norm layer is connected to the input end of the second adding unit through the MLP layer.

[0028] Preferably, each CFFEM module includes a third addition unit, a global average pooling layer, a first convolutional layer, a first activation layer, a second convolutional layer, a third convolutional layer, a first softmax unit, a second softmax unit, a first multiplication unit, a second multiplication unit and a fourth addition unit;

[0029] One input end of the third adding unit is connected to the front phase remote sensing image, and the other input end is connected to the back phase remote sensing image;

[0030] The output end of the third adding unit is connected to the input end of the first activation layer through the global average pooling layer, the first convolutional layer and the first activation layer in sequence;

[0031] The output end of the first activation layer is connected to the input end of the second convolutional layer and the input end of the third convolutional layer respectively;

[0032] The output end of the second convolutional layer is connected to an input end of the first multiplication unit through the first softmax unit;

[0033] Another input end of the first multiplication unit is connected to the remote sensing image of the previous phase;

[0034] The output end of the third convolutional layer is connected to an input end of the second multiplication unit through the second softmax unit;

[0035] Another input terminal of the second multiplication unit is connected to the later-phase remote sensing image;

[0036] The output end of the first multiplication unit and the output end of the second multiplication unit are both connected to the input end of the fourth addition unit;

[0037] The output end of the fourth adding unit is the output end of the CFFEM module.

[0038] Preferably, the improved change detection network model further includes a decoder;

[0039] The decoder includes a fifth CFFEM module, a sixth CFFEM module, a seventh CFFEM module and an eighth CFFEM module, a CPM module, a first upsampling module, a first ResidualBlock, a second upsampling module, a second ResidualBlock and a third Softmax unit;

[0040] The output end of the first CFFEM module is connected to the input end of the CPM module through the fifth CFFEM module;

[0041] The output end of the second CFFEM module is connected to the input end of the CPM module through the sixth CFFEM module;

[0042] The output end of the third CFFEM module is connected to the input end of the CPM module through the seventh CFFEM module;

[0043] The output end of the fourth CFFEM module is connected to the input end of the CPM module through the eighth CFFEM module;

[0044] The output end of the CPM module is connected to the input end of the third Softmax unit through the first up-sampling module, the first ResidualBlock, the second up-sampling module, and the second ResidualBlock in sequence;

[0045] The input end of the third Softmax unit is the output end of the decoder and also the output end of the improved change detection network model.

[0046] Preferably, the CPM module includes a fourth convolutional layer, a second activation layer, a batch normalization layer and a fifth convolutional layer connected in sequence.

[0047] Preferably, the two ResidualBlocks have the same structure, specifically including a first 3*3 convolutional layer, a Relu activation function, a second 3*3 convolutional layer and a third multiplication unit connected in sequence; the third multiplication unit is used to multiply the output of the second 3*3 convolutional layer by 0.1.

[0048] Preferably, during the training process of the improved change detection network model, the data sets used include the LEVIR-CD building change detection data set, the CLCD change detection data set and the SYSU-CD change detection data set.

[0049] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned remote sensing image change detection method when executing the computer program.

[0050] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a remote sensing image change detection method, which can achieve the following beneficial technical effects:

[0051] 1) The present invention can extract more detailed boundary information;

[0052] 2) The present invention can reduce the missed detection rate and false detection rate of the model;

[0053] 3) The present invention can improve the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0055] Figure 1 A schematic diagram of the structure of the improved change detection network model provided by the present invention;

[0056] Figure 2 A schematic diagram of the structure of the Shuffled-Sparse Attention module provided by the present invention;

[0057] Figure 3 A schematic diagram of the structure of the CFFEM module provided by the present invention;

[0058] Figure 4A schematic diagram of the structure of a decoder provided by the present invention;

[0059] Figure 5 A schematic diagram of the structure of the CPM module provided by the present invention;

[0060] Figure 6 A schematic diagram of the structure of the ResidualBlock provided by the present invention;

[0061] Figure 7 The present invention is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0063] The embodiment of the present invention discloses a remote sensing image change detection method, comprising the following steps:

[0064] Acquire the dual-temporal remote sensing image to be detected: wherein the dual-temporal remote sensing image to be detected includes the front-phase remote sensing image and the back-phase remote sensing image;

[0065] The dual-temporal remote sensing image is input into a pre-trained improved change detection network model to obtain the change area and change type of the dual-temporal remote sensing image; wherein the improved change detection network model introduces a Shuffled-Sparse Attention module and a CFFEM module.

[0066] Further, such as Figure 1 As shown, the improved change detection network model includes an encoder;

[0067] The encoder includes a first downsampling module, a first Shuffled-Sparse Attention module, a second downsampling module, a second Shuffled-Sparse Attention module, a third downsampling module, a third Shuffled-Sparse Attention module, a fourth downsampling module, and a fourth Shuffled-Sparse Attention module, which are sequentially connected;

[0068] The encoder further includes a fifth downsampling module, a fifth Shuffled-SparseAttention module, a sixth downsampling module, a sixth Shuffled-Sparse Attention module, a seventh downsampling module, a seventh Shuffled-Sparse Attention module, an eighth downsampling module, and an eighth Shuffled-SparseAttention module connected in sequence;

[0069] The input end of the first down-sampling module is connected to the remote sensing image of the previous phase;

[0070] The input end of the fifth down-sampling module is connected to the later phase remote sensing image;

[0071] The encoder also includes a first CFFEM module, a second CFFEM module, a third CFFEM module and a fourth CFFEM module;

[0072] The output end of the first Shuffled-Sparse Attention module and the output end of the fifth Shuffled-Sparse Attention module are both connected to the input end of the first CFFEM module;

[0073] The output end of the second Shuffled-Sparse Attention module and the output end of the sixth Shuffled-Sparse Attention module are both connected to the input end of the second CFFEM module;

[0074] The output end of the third Shuffled-Sparse Attention module and the output end of the seventh Shuffled-Sparse Attention module are both connected to the input end of the third CFFEM module;

[0075] The output end of the fourth Shuffled-Sparse Attention module and the output end of the eighth Shuffled-Sparse Attention module are both connected to the input end of the fourth CFFEM module.

[0076] Further, such as Figure 2 As shown, each Shuffled-Sparse Attention module includes a first Norm layer, a shuffled Attention Layer, a first addition unit, a second Norm layer, an MLP layer, and a second addition unit;

[0077] The input end of the first Norm layer is the input end of the Shuffled-Sparse Attention module;

[0078] The output end of the second adding unit is the output end of the Shuffled-Sparse Attention module;

[0079] An input end of the first Norm layer is connected to an input end of the first adding unit;

[0080] The output end of the first Norm layer is connected to the input end of the shuffled Attention Layer;

[0081] An output end of the shuffled Attention Layer is connected to an input end of the first adding unit;

[0082] The output end of the first adding unit is connected to the input end of the second Norm layer and the second adding unit respectively;

[0083] The output end of the second Norm layer is connected to the input end of the second adding unit through the MLP layer.

[0084] It is understandable that the Shuffled-Sparse Attention module introduces a sparse design and only focuses on some important areas in the input data, thereby significantly reducing the computational complexity and memory consumption. This sparse design makes the model more efficient when processing large-scale data sets. In tasks such as image processing, the local information of the data is crucial to capturing image features. The Shuffled-Sparse Attention module retains the two-dimensional geometry and locality of the data in its design, which helps to better capture the spatial relationship in the image. The Shuffled-Sparse Attention module randomly shuffles the order of the sequence data or feature maps through multiple iterations in each iteration, so that each element or region has the opportunity to gain attention in different contexts. This shuffling method increases the diversity of the data and helps the model capture more sequence information or image features. Since different attention weights and shuffling methods are used in each iteration, the Shuffled-Sparse Attention module can learn richer feature representations, thereby improving the generalization ability of the model. This enables the model to show better robustness when processing unseen data.

[0085] Further, such as Figure 3As shown, each CFFEM module includes a third addition unit, a global average pooling layer, a first convolutional layer, a first activation layer, a second convolutional layer, a third convolutional layer, a first softmax unit, a second softmax unit, a first multiplication unit, a second multiplication unit and a fourth addition unit;

[0086] One input end of the third adding unit is connected to the front phase remote sensing image, and the other input end is connected to the back phase remote sensing image;

[0087] The output end of the third adding unit is connected to the input end of the first activation layer through the global average pooling layer, the first convolutional layer and the first activation layer in sequence;

[0088] The output end of the first activation layer is connected to the input end of the second convolutional layer and the input end of the third convolutional layer respectively;

[0089] The output end of the second convolutional layer is connected to an input end of the first multiplication unit through the first softmax unit;

[0090] Another input end of the first multiplication unit is connected to the remote sensing image of the previous phase;

[0091] The output end of the third convolutional layer is connected to an input end of the second multiplication unit through the second softmax unit;

[0092] Another input terminal of the second multiplication unit is connected to the later-phase remote sensing image;

[0093] The output end of the first multiplication unit and the output end of the second multiplication unit are both connected to the input end of the fourth addition unit;

[0094] The output end of the fourth adding unit is the output end of the CFFEM module.

[0095] It should be noted that: the first activation layer adopts the Leaky Relu activation function.

[0096] Further, such as Figure 4 As shown, the improved change detection network model also includes a decoder;

[0097] The decoder includes a fifth CFFEM module, a sixth CFFEM module, a seventh CFFEM module and an eighth CFFEM module, a CPM module, a first upsampling module, a first ResidualBlock, a second upsampling module, a second ResidualBlock and a third Softmax unit;

[0098] The output end of the first CFFEM module is connected to the input end of the CPM module through the fifth CFFEM module;

[0099] The output end of the second CFFEM module is connected to the input end of the CPM module through the sixth CFFEM module;

[0100] The output end of the third CFFEM module is connected to the input end of the CPM module through the seventh CFFEM module;

[0101] The output end of the fourth CFFEM module is connected to the input end of the CPM module through the eighth CFFEM module;

[0102] The output end of the CPM module is connected to the input end of the third Softmax unit through the first up-sampling module, the first ResidualBlock, the second up-sampling module, and the second ResidualBlock in sequence;

[0103] The input end of the third Softmax unit is the output end of the decoder and also the output end of the improved change detection network model.

[0104] Further, such as Figure 5 As shown, the CPM module includes a fourth convolutional layer, a second activation layer, a batch normalization layer and a fifth convolutional layer connected in sequence.

[0105] It should be noted that the second activation layer adopts the Relu activation function.

[0106] Further, such as Figure 6 As shown, the two ResidualBlocks have the same structure, specifically including a first 3*3 convolutional layer, a Relu activation function, a second 3*3 convolutional layer and a third multiplication unit connected in sequence; the third multiplication unit is used to multiply the output of the second 3*3 convolutional layer by 0.1.

[0107] Preferably, during the training process of the improved change detection network model, the data sets used include the LEVIR-CD building change detection data set, the CLCD change detection data set and the SYSU-CD change detection data set.

[0108] It should be noted that:

[0109] The present invention adjusts the 256×256×3 dual-phase remote sensing image to 512×512×3 before inputting it into the encoder;

[0110] The decoder of the present invention outputs a prediction map of size 512×512*2, and the prediction map of 512×512*2 is adjusted to 256×256 for index score calculation.

[0111] During training, the present invention uses random scale cropping, Gaussian blur, random flipping, random rescaling and random color jittering to perform data enhancement on the dataset.

[0112] The present invention uses random initialization on 4 NVIDIA 3090 GPUs to train the improved change detection network model.

[0113] In the training process, the batch size is set to 16 and the initial learning rate is set to 4.1e -4 , the number of training rounds is set to 300.

[0114] The present invention uses linear decay to reduce the learning rate until the last stage.

[0115] In the inference stage, the present invention freezes the model weights, adjusts the input 256×256×3 dual-phase remote sensing image to 512×512×3, and then passes it to the improved change detection network model constructed by the present invention to obtain a 512×512×2 prediction map, and adjusts the 512×512×2 prediction map to the original spatial resolution of 256×256.

[0116] Optimizers are a method of updating parameters during network training to ensure that the loss function continues to decrease along the correct gradient direction until it reaches the global minimum.

[0117] The present invention selects Adam as the optimization algorithm during the training process. The Adam algorithm simultaneously calculates the first-order moment estimate and the second-order moment estimate in the t-th iteration, as shown in the following equations (1) and (2):

[0118] m t =β1.m t-1 +(1-β1).g t (1);

[0119] v t =β2.v t-1 +(1-β2).g t 2 (2);

[0120] Where m t and v t denote the first-order moment estimate and the second-order moment estimate in the t-th iteration, respectively, and m t-1 and v t-1 They represent the first-order moment estimate and the second-order moment estimate in the t-1th iteration, respectively, and g t represents the gradient of the tth iteration, β1 and β2 are the decay rates, usually taking values ​​of 0.9 and 0.99 respectively.

[0121] The first-order moment estimation and the second-order moment estimation are initialized to 0 at the beginning of the iteration, which may cause a large deviation. To avoid this initialization deviation, the present invention performs initialization on m t and v t The deviation correction formula is shown in equations (3) and (4):

[0122]

[0123] In the formula and is m t and v t The correction value of and are the t-th power of β1 and β2 respectively.

[0124] Finally, the network parameters are updated using the following formula (5):

[0125]

[0126] Where θ t-1 represents the network parameters in the t-1th iteration, θ t Indicates the network parameter in the tth iteration λ is the learning rate, ε represents a very small amount to prevent the denominator from being 0, and is generally set to 10 -8 .

[0127] Finally, as shown in Tables 1 to 3, the present invention compares the accuracy of different detection models on different data sets:

[0128] Table 1 Accuracy comparison of different models on the LEVIR-CD dataset

[0129]

[0130] Table 2 Accuracy comparison of different models on the CLCD dataset

[0131]

[0132] Table 3 Accuracy comparison of different models on the SYSU dataset

[0133]

[0134]

[0135] It is not difficult to see that:

[0136] For the CLCD dataset and LEVIR-CD dataset, the proposed method performs better than other models in terms of Precision (%), Recall (%), F1-score (%), IoU (%) and OA (%);

[0137] For the SYSU dataset, the proposed method is only slightly inferior to SNUNet in Precision (%), and is superior to other models in Recall (%), F1-score (%), IoU (%) and OA (%).

[0138] The embodiment of the present invention also discloses a physical structure diagram of an electronic device, such as Figure 7 As shown, the electronic device may include: a processor 701, a communication interface 702, a memory 703 and a communication bus 704, wherein the processor 701, the communication interface 702 and the memory 703 communicate with each other via the communication bus 704. The processor 701 may call the logic instructions in the memory 703 to execute the above remote sensing image change detection method.

[0139] In addition, the logic instructions in the above-mentioned memory 703 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in the embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0140] On the other hand, an embodiment of the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the above-mentioned remote sensing image change detection method.

[0141] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0142] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A remote sensing image change detection method, characterized in that: The following steps are involved: Acquire the dual-temporal remote sensing image to be detected: wherein the dual-temporal remote sensing image to be detected includes the front-phase remote sensing image and the back-phase remote sensing image; The dual-temporal remote sensing image is input into a pre-trained improved change detection network model to obtain the change area and change type of the dual-temporal remote sensing image; wherein the improved change detection network model introduces a Shuffled-SparseAttention module and a CFFEM module.

2. A remote sensing image change detection method according to claim 1, characterized in that: The improved change detection network model includes an encoder; The encoder includes a first downsampling module, a first Shuffled-Sparse Attention module, a second downsampling module, a second Shuffled-Sparse Attention module, a third downsampling module, a third Shuffled-Sparse Attention module, a fourth downsampling module, and a fourth Shuffled-Sparse Attention module, which are sequentially connected; The encoder further includes a fifth downsampling module, a fifth Shuffled-Sparse Attention module, a sixth downsampling module, a sixth Shuffled-Sparse Attention module, a seventh downsampling module, a seventh Shuffled-Sparse Attention module, an eighth downsampling module, and an eighth Shuffled-Sparse Attention module connected in sequence; The input end of the first down-sampling module is connected to the remote sensing image of the previous phase; The input end of the fifth down-sampling module is connected to the later phase remote sensing image; The encoder also includes a first CFFEM module, a second CFFEM module, a third CFFEM module and a fourth CFFEM module; The output end of the first Shuffled-Sparse Attention module and the output end of the fifth Shuffled-Sparse Attention module are both connected to the input end of the first CFFEM module; The output end of the second Shuffled-Sparse Attention module and the output end of the sixth Shuffled-Sparse Attention module are both connected to the input end of the second CFFEM module; The output end of the third Shuffled-Sparse Attention module and the output end of the seventh Shuffled-Sparse Attention module are both connected to the input end of the third CFFEM module; The output end of the fourth Shuffled-Sparse Attention module and the output end of the eighth Shuffled-Sparse Attention module are both connected to the input end of the fourth CFFEM module.

3. A remote sensing image change detection method according to claim 2, characterized in that: Each Shuffled-Sparse Attention module includes the first Norm layer, the shuffled Attention Layer, the first addition unit, the second Norm layer, the MLP layer and the second addition unit; The input end of the first Norm layer is the input end of the Shuffled-Sparse Attention module; The output end of the second adding unit is the output end of the Shuffled-Sparse Attention module; An input end of the first Norm layer is connected to an input end of the first adding unit; The output end of the first Norm layer is connected to the input end of the shuffled Attention Layer; An output end of the shuffled Attention Layer is connected to an input end of the first adding unit; The output end of the first adding unit is connected to the input end of the second Norm layer and the second adding unit respectively; The output end of the second Norm layer is connected to the input end of the second adding unit through the MLP layer.

4. A remote sensing image change detection method according to claim 3, characterized in that: Each CFFEM module includes a third addition unit, a global average pooling layer, a first convolutional layer, a first activation layer, a second convolutional layer, a third convolutional layer, a first softmax unit, a second softmax unit, a first multiplication unit, a second multiplication unit, and a fourth addition unit; One input end of the third adding unit is connected to the front phase remote sensing image, and the other input end is connected to the back phase remote sensing image; The output end of the third adding unit is connected to the input end of the first activation layer through the global average pooling layer, the first convolutional layer and the first activation layer in sequence; The output end of the first activation layer is connected to the input end of the second convolutional layer and the input end of the third convolutional layer respectively; The output end of the second convolutional layer is connected to an input end of the first multiplication unit through the first softmax unit; Another input end of the first multiplication unit is connected to the remote sensing image of the previous phase; The output end of the third convolutional layer is connected to an input end of the second multiplication unit through the second softmax unit; Another input terminal of the second multiplication unit is connected to the later-phase remote sensing image; The output end of the first multiplication unit and the output end of the second multiplication unit are both connected to the input end of the fourth addition unit; The output end of the fourth adding unit is the output end of the CFFEM module.

5. A remote sensing image change detection method according to claim 4, characterized in that: The improved change detection network model also includes a decoder; The decoder includes a fifth CFFEM module, a sixth CFFEM module, a seventh CFFEM module and an eighth CFFEM module, a CPM module, a first upsampling module, a first ResidualBlock, a second upsampling module, a second ResidualBlock and a third Softmax unit; The output end of the first CFFEM module is connected to the input end of the CPM module through the fifth CFFEM module; The output end of the second CFFEM module is connected to the input end of the CPM module through the sixth CFFEM module; The output end of the third CFFEM module is connected to the input end of the CPM module through the seventh CFFEM module; The output end of the fourth CFFEM module is connected to the input end of the CPM module through the eighth CFFEM module; The output end of the CPM module is connected to the input end of the third Softmax unit through the first up-sampling module, the first ResidualBlock, the second up-sampling module, and the second ResidualBlock in sequence; The input end of the third Softmax unit is the output end of the decoder and also the output end of the improved change detection network model.

6. A remote sensing image change detection method according to claim 5, characterized in that: The CPM module includes a fourth convolutional layer, a second activation layer, a batch normalization layer and a fifth convolutional layer which are connected in sequence.

7. A remote sensing image change detection method according to claim 5, characterized in that: The two ResidualBlocks have the same structure, specifically including a first 3*3 convolutional layer, a Relu activation function, a second 3*3 convolutional layer and a third multiplication unit connected in sequence; the third multiplication unit is used to multiply the output of the second 3*3 convolutional layer by 0.

1.

8. The remote sensing image change detection method according to claim 1, characterized in that: During the training process of the improved change detection network model, the data sets used include the LEVIR-CD building change detection data set, the CLCD change detection data set and the SYSU-CD change detection data set.

9. An electronic device 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 remote sensing image change detection method according to any one of claims 1 to 8 is implemented.

Images