A method and device for detecting changes in synthetic aperture radar images

Abstract

The present invention discloses a synthetic aperture radar image change detection method and device, the method comprising: obtaining a dual-phase SAR image training data set; the dual-phase SAR image training data set comprises a dual-phase SAR image and corresponding label information; the dual-phase SAR image is two SAR images acquired at the same position at different times; using the preprocessed dual-phase SAR image training data set, training a preset image change detection model to obtain the trained image change detection model; using the trained image change detection model, processing the acquired dual-phase SAR image to obtain an image change detection result.

Description

Technical Field

[0001] The invention relates to the fields of remote sensing and artificial intelligence, and in particular to a synthetic aperture radar image change detection method and device. Background Art

[0002] At present, the task of remote sensing image change detection is to detect the change information in remote sensing images of the same area at different phases. Due to the influence of human activities and ecosystems, the surface of the earth is changing all the time. Timely and accurate acquisition of surface change information can provide technical support for various decisions, such as environmental protection, disaster detection and natural resource management. With the development of remote sensing technology, high-resolution visible light image data, multispectral image data, hyperspectral image data and SAR image data are gradually being adopted in the field of change detection. SAR images are generated by synthetic aperture radar systems. They emit and receive surface radar waves through active imaging. They are an active earth observation system. Due to the unique imaging method of SAR images, the obtained images are not affected by factors such as light and atmosphere.

[0003] At present, the change detection method based on convolutional neural network (CNN) performs well in extracting local information, but it is difficult to extract global information of high-resolution images. However, for change detection, the feature interaction between dual-phase images is crucial. How to achieve efficient and accurate SAR image change detection based on the feature interaction between dual-phase images is an urgent problem to be solved. Summary of the invention

[0004] The present invention mainly solves the problem of how to realize efficient and accurate SAR image change detection based on feature interaction between dual-phase images. The present invention discloses a synthetic aperture radar image change detection method and device.

[0005] In a first aspect of an embodiment of the present application, a synthetic aperture radar image change detection method is disclosed, comprising:

[0006] S1, obtaining a dual-phase SAR image training data set; the dual-phase SAR image training data set includes dual-phase SAR images and corresponding label information; the dual-phase SAR image is a SAR image acquired at the same location at two different times;

[0007] S2, using the preprocessed dual-phase SAR image training data set to train a preset image change detection model to obtain a trained image change detection model;

[0008] S3, using the trained image change detection model, processing the acquired dual-phase SAR image to obtain an image change detection result.

[0009] The preprocessing process of the dual-phase SAR image training data set includes:

[0010] S21, performing outlier elimination processing on the dual-phase SAR image training data set to obtain a first information set;

[0011] S22: Perform boundary check processing on the first information set to obtain a preprocessed dual-phase SAR image training data set.

[0012] The performing boundary check processing on the first information set to obtain a preprocessed dual-phase SAR image training data set includes:

[0013] S221, for each data attribute of the first information set, preset a corresponding value range; the value range includes an upper limit value of the value range and a lower limit value of the value range;

[0014] S222, for each data of each data attribute of the first information set, judging whether the value of the data is within the value range according to the value range of the data attribute; if the data is within the value range, not processing the data; if the data is not within the value range, setting the value of the data to the upper limit value or the lower limit value of the value range that is closest to the data;

[0015] S223, after completing S222 for all data attributes, a preprocessed dual-phase SAR image training data set is obtained.

[0016] The image change detection model includes: a local feature extraction module, a global feature extraction module, a space exchange module, a channel exchange module, a channel mixing module and a feature classifier;

[0017] The local feature extraction module is connected to the global feature extraction module and is used to extract local features from the input dual-phase SAR image to obtain local feature information;

[0018] The global feature extraction module is connected to the space exchange module and is used to extract global features from the input local feature information to obtain global feature information;

[0019] The spatial exchange module is connected to the channel exchange module and is used to perform data exchange of spatial dimensions on the input global feature information to obtain spatial exchange data information;

[0020] The channel exchange module is connected to the channel mixing module and is used to perform channel dimension data exchange on the input spatial exchange data information to obtain channel exchange data information;

[0021] The channel mixing module is connected to the feature classifier and is used to perform channel-dimensional data cross-mixing on the input channel exchange data information to obtain channel mixed data information;

[0022] The feature classifier is used to perform feature classification processing on the channel mixed data information to obtain the image change detection result, including a convolution layer, a regularization layer and an activation function layer; the convolution layer includes three convolution modules connected in series; the input end of the convolution layer is used to receive the channel mixed data information; the output end of the convolution layer is connected to the input end of the regularization layer; the output end of the regularization layer is connected to the input end of the activation function layer; the activation function layer is used to output the image change detection result.

[0023] The global feature extraction module includes two parallel global feature extraction submodules; the input ends of the two parallel global feature extraction submodules are respectively connected to the output ends of the local feature extraction module; the output ends of the two parallel global feature extraction submodules are respectively connected to the input end of the space exchange module;

[0024] The global feature extraction submodule includes an input layer, a first adder, a position encoder, a first normalization layer, a multi-head self-attention layer, a second adder, a second normalization layer, a multi-layer perceptron, a third adder, and an output layer;

[0025] The input layer is used to receive and obtain local feature information; the first adder and the position encoder are connected to the input layer respectively;

[0026] The position encoder is used to extract the position of the local feature information to obtain the position information;

[0027] The first adder is connected to the position encoder and is used to perform addition processing on the local feature information and the position information; the first input end of the first adder is connected to the output end of the input layer; the second input end of the first adder is connected to the output end of the position encoder;

[0028] The first normalization layer is connected to the multi-head self-attention layer; the second normalization layer is connected to the multi-layer perceptron;

[0029] The first output terminal of the first adder is connected to the input terminal of the first normalization layer; the second output terminal of the first adder is connected to the first input terminal of the second adder;

[0030] An input end of the multi-head self-attention layer is connected to a second input end of the second adder;

[0031] The first output terminal of the second adder is connected to the input terminal of the second normalization layer; the second output terminal of the second adder is connected to the first input terminal of the third adder;

[0032] The output end of the multilayer perceptron is connected to the second input end of the third adder;

[0033] The output end of the third adder is connected to the input end of the output layer.

[0034] The training process of the image change detection model includes:

[0035] Initialize the number of training iterations;

[0036] Using the dual-phase SAR images in the dual-phase SAR image training data set as input data to the image change detection model; using label information corresponding to the dual-phase SAR images input to the image change detection model as label data;

[0037] Using the image change detection model, the input data is processed to obtain an image change detection result;

[0038] Performing difference calculation processing on the obtained image change detection result and label data to obtain a difference value;

[0039] Determine whether the difference value satisfies a convergence condition, and obtain a first determination result;

[0040] When the first judgment result is no, judging whether the training iteration number value is equal to a preset training number threshold, and obtaining a second judgment result;

[0041] When the second judgment result is no, determining that the model training state does not meet the training termination condition;

[0042] When the second judgment result is yes, determining that the model training state satisfies the training termination condition;

[0043] When the first judgment result is yes, determining that the model training state satisfies a training termination condition;

[0044] When the model training state does not meet the training termination condition, the image change detection model is updated with parameters using a parameter update model, the training iteration count is increased by 1, and the execution is triggered to input the dual-phase SAR image in the dual-phase SAR image training data set as input data into the image change detection model;

[0045] When the model training state satisfies the training termination condition, the training process of the image change detection model is completed to obtain a trained image change detection model.

[0046] The parameter updating model is:

[0047]

[0048] θ←θ+v;

[0049] In the formula, is the loss function of the i-th dual-phase SAR image in the dual-phase SAR image training data set, v is the parameter update value, θ is the parameter of the image change detection model, η is the initial parameter learning rate, α is the momentum angle parameter, 0≤α≤π / 4, It means to find partial derivatives with respect to variable θ; η and α are preset values.

[0050] In a second aspect of an embodiment of the present application, a synthetic aperture radar image change detection device is disclosed, the device comprising:

[0051] A memory storing executable program code;

[0052] a processor coupled to the memory;

[0053] The processor calls the executable program code stored in the memory to execute the synthetic aperture radar image change detection method.

[0054] In a third aspect of an embodiment of the present application, a computer storable medium is disclosed, wherein the computer storable medium stores computer instructions, and when the computer instructions are called, they are used to execute the synthetic aperture radar image change detection method.

[0055] In a fourth aspect of the embodiments of the present application, an information data processing terminal is disclosed, and the information data processing terminal is used to implement the synthetic aperture radar image change detection method.

[0056] The beneficial effects of the present invention are:

[0057] The method is aimed at the field of change detection of SAR image data. Most of the current change detection methods do not consider the local features and global features of dual-phase images at the same time. The present invention proposes a method of combining a network with pre-trained weights and a global feature extraction module network to extract local features and global features of the input image. In addition, the essence of the change detection task is to detect changes, so the interaction between dual-phase features is particularly important. To address this phenomenon, the present invention interacts the features of dual-phase input through three different feature interaction modules. The experimental results on two data sets show that the comprehensive evaluation index of the present invention is higher than that of the current advanced change detection method, which also verifies the effectiveness of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 1 is a flow chart for implementing the method of the present invention;

[0059] Figure 2 It is a composition diagram of the device of the present invention;

[0060] Figure 3 It is a composition diagram of the global feature extraction module of the present invention;

[0061] Figure 4 It is a schematic diagram of the data exchange process of the space exchange module and the channel exchange module of the present invention. DETAILED DESCRIPTION

[0062] In order to better understand the content of the present invention, an embodiment is given here.

[0063] Figure 1 1 is a flow chart for implementing the method of the present invention; Figure 2 It is a composition diagram of the device of the present invention; Figure 3 It is a composition diagram of the global feature extraction module of the present invention; Figure 4 It is a schematic diagram of the data exchange process of the space exchange module and the channel exchange module of the present invention.

[0064] The essence of the change detection task is to detect changes in the same position in different phases. Currently, most change detection methods are committed to segmenting the semantics of a single phase and detecting changes in the segmented results on this basis. However, this ignores the interactive features of dual-phase images to a certain extent. Therefore, in order to further improve the results of image change detection, the present invention proposes a SAR image change detection method based on feature interaction of a global feature extraction module.

[0065] In a first aspect of an embodiment of the present application, a synthetic aperture radar image change detection method is disclosed, comprising:

[0066] S1, obtaining a dual-phase SAR image training data set; the dual-phase SAR image training data set includes dual-phase SAR images and corresponding label information; the dual-phase SAR image is two SAR images acquired at the same location at different times;

[0067] S2, using the preprocessed dual-phase SAR image training data set to train a preset image change detection model to obtain a trained image change detection model;

[0068] S3, using the trained image change detection model, processing the acquired dual-phase SAR image to obtain an image change detection result.

[0069] The preprocessing process of the dual-phase SAR image training data set includes:

[0070] S21, performing outlier elimination processing on the task requirement information and the task assignment information to obtain a first information set;

[0071] S22: Perform boundary check processing on the first information set to obtain a preprocessed dual-phase SAR image training data set.

[0072] The performing boundary check processing on the first information set to obtain a preprocessed dual-phase SAR image training data set includes:

[0073] S221, for each data attribute of the first information set, preset a corresponding value range; the value range includes an upper limit value of the value range and a lower limit value of the value range;

[0074] S222, for each data of each data attribute of the first information set, judging whether the value of the data is within the value range according to the value range of the data attribute; if the data is within the value range, not processing the data; if the data is not within the value range, setting the value of the data to the upper limit value or the lower limit value of the value range that is closest to the data;

[0075] S223, after completing S222 for all data attributes, a preprocessed dual-phase SAR image training data set is obtained.

[0076] The image change detection model includes: a local feature extraction module, a global feature extraction module, a space exchange module, a channel exchange module and a channel mixing module, and a feature classifier;

[0077] The local feature extraction module is connected to the global feature extraction module and is used to extract local features from the input dual-phase SAR image to obtain local feature information;

[0078] The global feature extraction module is connected to the space exchange module and is used to extract global features from the input local feature information to obtain global feature information;

[0079] The spatial exchange module is connected to the channel exchange module and is used to perform data exchange of spatial dimensions on the input global feature information to obtain spatial exchange data information;

[0080] The channel exchange module is connected to the channel mixing module and is used to perform channel dimension data exchange on the input spatial exchange data information to obtain channel exchange data information;

[0081] The channel mixing module is connected to the feature classifier and is used to perform channel-dimensional data cross-mixing on the input channel exchange data information to obtain channel mixed data information;

[0082] The feature classifier is used to perform feature classification processing on the channel mixed data information to obtain the image change detection result, including a convolution layer, a regularization layer and an activation function layer, which are used to perform feature classification processing on the channel mixed data information to obtain the image change detection result; the convolution layer includes three convolution modules connected in series; the input end of the convolution layer is used to receive the channel mixed data information; the output end of the convolution layer is connected to the input end of the regularization layer; the output end of the regularization layer is connected to the input end of the activation function layer. The activation function layer is used to output the image change detection result.

[0083] The global feature extraction module includes two parallel global feature extraction submodules; the input ends of the two parallel global feature extraction submodules are respectively connected to the output ends of the local feature extraction module; the output ends of the two parallel global feature extraction submodules are respectively connected to the input end of the space exchange module;

[0084] The global feature extraction submodule includes an input layer, a first adder, a position encoder, a first normalization layer, a multi-head self-attention layer, a second adder, a second normalization layer, a multi-layer perceptron, a third adder, and an output layer;

[0085] The input layer is used to receive and obtain local feature information; the first adder and the position encoder are connected to the input layer respectively;

[0086] The position encoder is used to extract the position of the local feature information to obtain the position information; the position information is the coordinate information of the local feature;

[0087] The first adder is connected to the position encoder and is used to perform addition processing on the local feature information and the position information;

[0088] The first normalization layer is connected to the multi-head self-attention layer; the second normalization layer is connected to the multi-layer perceptron;

[0089] The first output terminal of the first adder is connected to the input terminal of the first normalization layer; the second output terminal of the first adder is connected to the first input terminal of the second adder;

[0090] An input end of the multi-head self-attention layer is connected to a second input end of the second adder;

[0091] The first output terminal of the second adder is connected to the input terminal of the second normalization layer; the second output terminal of the second adder is connected to the first input terminal of the third adder;

[0092] The output end of the multilayer perceptron is connected to the second input end of the third adder;

[0093] The output end of the third adder is connected to the input end of the output layer.

[0094] The training process of the image change detection model includes:

[0095] Initialize the number of training iterations;

[0096] Using the dual-phase SAR images in the dual-phase SAR image training data set as input data to the image change detection model; using label information corresponding to the dual-phase SAR images input to the image change detection model as label data;

[0097] Using the image change detection model, the input data is processed to obtain an image change detection result;

[0098] Performing difference calculation processing on the obtained image change detection result and label data to obtain a difference value;

[0099] Determine whether the difference value satisfies a convergence condition, and obtain a first determination result;

[0100] When the first judgment result is no, judging whether the training iteration number value is equal to the training number threshold, and obtaining a second judgment result;

[0101] When the second judgment result is no, determining that the model training state does not meet the training termination condition;

[0102] When the second judgment result is yes, determining that the model training state satisfies the training termination condition;

[0103] When the first judgment result is yes, determining that the model training state satisfies a training termination condition;

[0104] When the model training state does not meet the training termination condition, the image change detection model is updated with parameters using a parameter update model, the training iteration count is increased by 1, and the execution is triggered to input the dual-phase SAR image in the dual-phase SAR image training data set as input data into the image change detection model;

[0105] When the model training state satisfies the training termination condition, the training process of the image change detection model is completed to obtain a trained image change detection model.

[0106] The parameter updating model is:

[0107]

[0108] θ←θ+v;

[0109] In the formula, is the loss function of the i-th dual-phase SAR image in the dual-phase SAR image training data set, v is the parameter update value, θ is the parameter of the image change detection model, η is the initial parameter learning rate, α is the momentum angle parameter, 0≤α≤π / 4, It means to find partial derivatives with respect to variable θ; η and α are preset values.

[0110] The image change detection result is used to indicate whether the position information of the acquired dual-phase SAR image data has changed. A value of 1 indicates a change, and a value of 0 indicates no change.

[0111] The difference value satisfies the convergence condition, which means that the difference value is less than a preset convergence threshold; the difference value does not satisfy the convergence condition, which means that the difference value is not less than the preset convergence threshold.

[0112] The data exchange of the spatial dimension of the spatial exchange module is to input the two data of the spatial exchange module and Extract the spatial dimension matrix, including and use replace In use replace In Then extract the spatial dimension matrix, including and use replace In use replace In The process is as Figure 4 The left side of the exchange process.

[0113] The channel dimension data exchange of the channel exchange module is to input the two data of the spatial exchange module and Extract the channel dimension matrix, including and use replace In use replace In The process is as Figure 4 The right side of the exchange process.

[0114] The channel dimension data cross-mixing of the channel mixing module is to and Fusion to obtain new data in

[0115] Spatial exchange module: For the output of the previous module, the features are exchanged in the spatial domain. This spatial exchange module has no learnable parameters, but chooses a simple exchange strategy to exchange the features of the dual-phase image in the spatial dimension. The specific exchange method is as follows: Figure 4 As shown, the result is then upsampled by 2 times to get the output and The upsampling formula is as follows:

[0116]

[0117] Channel exchange module: The features of the dual-phase images are exchanged in the dimension of the channel. This spatial exchange module has no learnable parameters, but chooses a simple exchange strategy to exchange the features of the dual-phase images in the dimension of the channel. The specific exchange method is as follows: Figure 4 As shown, the result is then upsampled by 2 times to get the output and The upsampling formula is as follows:

[0118]

[0119] Channel mixing module: The features of the two-phase image are mixed in the channel dimension to generate features that are convenient for change detection tasks. The two-phase input is mixed in the channel dimension. By cross-mixing the features of the two phases, the features of the two phases are aligned one by one in position, which can better aggregate the change information. The result is then upsampled by 2 times to obtain the output The specific formula is as follows:

[0120] Mix(X k ,Y k )=Conv2D(stack(X k ,Y k ),ch in =2C k ,ch out =C k ,group=C k )

[0121] in, stack(*,*) is a simple stacking operation.

[0122] The feature classifier is a pixel-level feature classifier that classifies the feature map with the interaction information of the two-phase image. The feature classifier consists of three convolutional layers, each of which has a convolution kernel of 3, followed by a regularization layer and an activation function PReLU. The activation function of the last layer is changed to a Sigmoid layer to detect the changed and non-changed areas. Finally, the threshold of Sigmoid is set to 0.5 to obtain the final result image F. out ∈R H×W , and the detection result of whether there is a transformation. If there is a transformation, the detection result is 1, and if there is no transformation, the detection result is 0.

[0123] The global feature extraction module is used to obtain the global features of the dual-phase SAR image: the dual-phase image is output from the Efficient_v4 network and Through a global feature extraction submodule, H0, W0 and C0 are the width, height and number of channels of the output feature map of the Efficient_v4 network respectively, and the global features of the image are obtained by using the self-attention mechanism of the global feature extraction submodule. The specific structure of the global feature extraction submodule is as follows Figure 3 As shown in the figure, the global feature extraction submodule adopts the self-attention mechanism. First, flatten it in the channel dimension, that is, Then add a positional encoding to extract the position of each semantic tag, then add them together, and then pass through the first normalization layer and the multi-head self-attention layer to implement the multi-head self-attention mechanism. For each tag, three different matrix weights are multiplied with it to obtain the sequence Q, key K and value V, as shown in the formula:

[0124] Q=F i 'W q ,K=F i 'W k ,V=F i 'W v

[0125] Among them, W v ,W q ,W k ,W v ∈R C×d are three learnable linear change parameters, and d is the number of channels after transformation. After obtaining the semantically labeled Q, K, and V, the self-attention weight can be obtained through self-attention calculation. The formula is as follows:

[0126]

[0127] In order to enrich the extracted feature information, the global feature extraction module adopts a multi-head self-attention mechanism, the core of which is to calculate multiple attention layers in parallel. The specific formula is as follows:

[0128] Multihead(Q,K,V)=Concat(head1,head2,...head h )W O

[0129] head j =Attention(F i 'W j q ,F i 'W j k ,F i 'W j v )

[0130] Among them, h refers to the number of attention heads, W j q ,W j k ,W j v ∈R C×d and W O ∈R hd×C is a linear learnable matrix parameter.

[0131] Then it is sent to a normalization layer, and then input into a multi-layer perceptron to obtain global semantic information. The multi-layer perceptron includes two linear fully connected layers and a Gaussian activation function layer set between the two linear fully connected layers. The specific formula is as shown in:

[0132] MLP(F′)=GELU(F′W1)W2

[0133] Where W1∈R C×2C ,W2∈R 2C×C is a learnable linear parameter matrix; GELU() represents a Gaussian activation function layer; F′ represents the input amount.

[0134] The activation function can be implemented using a Sigmoid function.

[0135] The local feature extraction module is implemented by connecting a convolution module with the first four layers of the Efficient_v4 network; specifically, the convolution module can use a pixel-level convolution layer, which sets the number of channels of the output image to the number of channels suitable for the input of the Efficient_v4 network, that is, the number of channels of visible light is 3.

[0136] For the input of the two-phase image I1∈RH×W×1 and I2∈R H×W×1 , where H and W are the width and height of the input image, respectively. The local feature extraction module, for the image structure of the SAR image, first sets the number of channels of the input image to the number of channels suitable for the input of the Efficient_v4 network, that is, the number of channels of visible light is 3 through a pixel-level convolution layer. Then it is passed through the first 4 layers of the Efficient_v4 network pre-trained on the ImageNet1000 dataset to obtain its local features. In this process, the dual-phase image is downsampled 8 times.

[0137] The performing boundary check processing on the first information set to obtain a preprocessed dual-phase SAR image training data set includes:

[0138] S221, for each type of data attribute of the first information set, the data collection information of the data is used as a known independent variable, and the data value of the data is used as a known dependent variable. The known independent variable and the known dependent variable are used to construct a curve to be approximated, and the function approximation method is used to perform curve fitting on the curve to be approximated to obtain the best consistent approximation polynomial f(Ix) of the type of data attribute; the best consistent approximation polynomial f(Ix) is used to calculate and process the known independent variables to obtain an approximate dependent variable; it is determined whether the absolute value of the difference between the approximate dependent variable and the corresponding known dependent variable is greater than a set second regression discrimination threshold; if it is greater than the second regression discrimination threshold, the data is deleted from the first information set; if it is less than the second regression discrimination threshold, the data is not processed;

[0139] S222: Execute S221 on the data of all data attributes of the first information set to obtain a preprocessed dual-phase SAR image training data set.

[0140] The curve fitting of the curve to be approximated by the function approximation method may be performed by using the best consistent linear approximation method. The best consistent approximation polynomial f(Ix) is expressed as:

[0141] f(Ix)=α P1 (Ix) P1 +α P1-1 (Ix) P1-1 +…+α2(Ix) 2 +α1(Ix)+α0,

[0142] Wherein, P1 is the order of the best consistent approximation polynomial f(Ix), α0, α1, α2, …, α P1 are the coefficients of the optimal consistent approximation polynomial f(Ix);

[0143] The data collection information includes collection time information, collection space information, etc.

[0144] The outlier elimination includes filling missing values, smoothing noise data, smoothing or deleting outlier points or normalization processing; the smoothed noise data is first determined to obtain noise data, and then the noise data is smoothed according to the data before and after the noise data; the noise data is a value whose value is less than the detection sensitivity of the sensor of the observed data, or greater than the measurement upper limit of the sensor of the observed data. The determination of the outlier point can adopt the Kalman filtering method. The filling value of the missing value can be determined by averaging the measured values ​​within a certain sampling interval before and after the missing value.

[0145] The normalization process is to map data in different value ranges to a specified value range, and its mathematical expression is:

[0146]

[0147] Among them, t is a data value in the data set, t max is the maximum value of all data in the data set, t min is the minimum value of all data in the data set, t 归一化 is the value of data t after normalization;

[0148] The category discrimination process is to discriminate the attributes of each data in each category of information in the normalized information set to see whether it is within the preset attribute range of the information, and to remove data that is not within the attribute range from the information.

[0149] In a second aspect of an embodiment of the present application, a synthetic aperture radar image change detection device is disclosed, the device comprising:

[0150] A memory storing executable program code;

[0151] a processor coupled to the memory;

[0152] The processor calls the executable program code stored in the memory to execute the synthetic aperture radar image change detection method.

[0153] In a third aspect of an embodiment of the present application, a computer storable medium is disclosed, wherein the computer storable medium stores computer instructions, and when the computer instructions are called by a computer, they are used to execute the synthetic aperture radar image change detection method.

[0154] In a fourth aspect of the embodiments of the present application, an information data processing terminal is disclosed, and the information data processing terminal is used to implement the synthetic aperture radar image change detection method.

[0155] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A synthetic aperture radar image change detection method, characterized in that: include: S1, obtaining a dual-phase SAR image training data set; the dual-phase SAR image training data set includes dual-phase SAR images and corresponding label information; the dual-phase SAR image is a SAR image acquired at the same location at two different times; S2, using the preprocessed dual-phase SAR image training data set to train a preset image change detection model to obtain a trained image change detection model; S3, using the trained image change detection model to process the acquired dual-phase SAR image to obtain an image change detection result; The preprocessing process of the dual-phase SAR image training data set includes: S21, performing outlier elimination processing on the dual-phase SAR image training data set to obtain a first information set; S22, performing boundary check processing on the first information set to obtain a preprocessed dual-phase SAR image training data set; The performing boundary check processing on the first information set to obtain a preprocessed dual-phase SAR image training data set includes: S221, for each data attribute of the first information set, preset a corresponding value range; the value range includes an upper limit value of the value range and a lower limit value of the value range; S222, for each data of each data attribute of the first information set, judging whether the value of the data is within the value range according to the value range of the data attribute; if the data is within the value range, not processing the data; if the data is not within the value range, setting the value of the data to the upper limit value or the lower limit value of the value range that is closest to the data; S223, after completing S222 for all data attributes, a preprocessed dual-phase SAR image training data set is obtained; The image change detection model includes: a local feature extraction module, a global feature extraction module, a space exchange module, a channel exchange module, a channel mixing module and a feature classifier; The local feature extraction module is connected to the global feature extraction module and is used to extract local features from the input dual-phase SAR image to obtain local feature information; The global feature extraction module is connected to the space exchange module and is used to extract global features from the input local feature information to obtain global feature information; The spatial exchange module is connected to the channel exchange module and is used to perform data exchange of spatial dimensions on the input global feature information to obtain spatial exchange data information; The channel exchange module is connected to the channel mixing module and is used to perform channel dimension data exchange on the input spatial exchange data information to obtain channel exchange data information; The channel mixing module is connected to the feature classifier and is used to perform channel-dimensional data cross-mixing on the input channel exchange data information to obtain channel mixed data information; The feature classifier is used to perform feature classification processing on the channel mixed data information to obtain the image change detection result, including a convolution layer, a regularization layer and an activation function layer; the convolution layer includes three convolution modules connected in series; the input end of the convolution layer is used to receive the channel mixed data information; the output end of the convolution layer is connected to the input end of the regularization layer; the output end of the regularization layer is connected to the input end of the activation function layer; the activation function layer is used to output the image change detection result.

2. The synthetic aperture radar image change detection method according to claim 1, characterized in that: The global feature extraction module includes two parallel global feature extraction submodules; the input ends of the two parallel global feature extraction submodules are respectively connected to the output ends of the local feature extraction module; the output ends of the two parallel global feature extraction submodules are respectively connected to the input end of the space exchange module; The global feature extraction submodule includes an input layer, a first adder, a position encoder, a first normalization layer, a multi-head self-attention layer, a second adder, a second normalization layer, a multi-layer perceptron, a third adder, and an output layer; The input layer is used to receive and obtain local feature information; the first adder and the position encoder are connected to the input layer respectively; The position encoder is used to extract the position of the local feature information to obtain the position information; The first adder is connected to the position encoder and is used to perform addition processing on the local feature information and the position information; the first input end of the first adder is connected to the output end of the input layer; the second input end of the first adder is connected to the output end of the position encoder; The first normalization layer is connected to the multi-head self-attention layer; the second normalization layer is connected to the multi-layer perceptron; The first output terminal of the first adder is connected to the input terminal of the first normalization layer; the second output terminal of the first adder is connected to the first input terminal of the second adder; An input end of the multi-head self-attention layer is connected to a second input end of the second adder; The first output terminal of the second adder is connected to the input terminal of the second normalization layer; the second output terminal of the second adder is connected to the first input terminal of the third adder; The output end of the multilayer perceptron is connected to the second input end of the third adder; The output end of the third adder is connected to the input end of the output layer.

3. The synthetic aperture radar image change detection method according to claim 2, characterized in that: The training process of the image change detection model includes: Initialize the number of training iterations; Using the dual-phase SAR images in the dual-phase SAR image training data set as input data to the image change detection model; using label information corresponding to the dual-phase SAR images input to the image change detection model as label data; Using the image change detection model, the input data is processed to obtain an image change detection result; Performing difference calculation processing on the obtained image change detection result and label data to obtain a difference value; Determine whether the difference value satisfies a convergence condition, and obtain a first determination result; When the first judgment result is no, judging whether the training iteration number value is equal to a preset training number threshold, and obtaining a second judgment result; When the second judgment result is no, determining that the model training state does not meet the training termination condition; When the second judgment result is yes, determining that the model training state satisfies the training termination condition; When the first judgment result is yes, determining that the model training state satisfies a training termination condition; When the model training state does not meet the training termination condition, the image change detection model is updated with parameters using a parameter update model, the training iteration count is increased by 1, and the execution is triggered to input the dual-phase SAR image in the dual-phase SAR image training data set as input data into the image change detection model; When the model training state satisfies the training termination condition, the training process of the image change detection model is completed to obtain a trained image change detection model.

4. The synthetic aperture radar image change detection method according to claim 3, characterized in that: The parameter updating model is: θ←θ+v; In the formula, is the loss function of the i-th dual-phase SAR image in the dual-phase SAR image training data set, v is the parameter update value, θ is the parameter of the image change detection model, η is the initial parameter learning rate, α is the momentum angle parameter, 0≤α≤π / 4, It means to find partial derivatives with respect to variable θ; η and α are preset values.

5. A synthetic aperture radar image change detection device, characterized in that: The device comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the synthetic aperture radar image change detection method according to any one of claims 1 to 4.

6. A computer storable medium, characterized in that: The computer storable medium stores computer instructions, and when the computer instructions are called by a computer, they are used to execute the synthetic aperture radar image change detection method according to any one of claims 1 to 4.

7. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the synthetic aperture radar image change detection method according to any one of claims 1 to 4.

Images