Single-polarization synthetic aperture radar change detection method and device

Abstract

The invention relates to the technical field of radar remote sensing image processing, in particular to a single-polarization synthetic aperture radar change detection method and device.The method comprises the steps that initial single-polarization SAR images of front and back time phases are preprocessed to obtain a final image, a multi-scale feature difference chart is calculated, pseudo tags are generated through fuzzy C-means clustering, and the final image is obtained; a UNet coding and decoding structure is used as a basic framework, a double-branch convolutional neural network structure is designed, an input layer, a convolutional layer, a pooling layer, a full connection layer and an output layer are adopted to form a twin convolutional neural network, residual connection is established between an input tensor and an output tensor after network processing, a change detection model is generated, and a change detection result is obtained. And outputting a detection result of the single-polarization synthetic aperture radar by using the change detection model. Therefore, the problems of low accuracy of a change monitoring result and the like caused by the defects that the processing capacity of speckle noise in an SAR image monitoring method is limited and a static threshold segmentation method is difficult to adapt to the global situation in the prior art are solved.

Description

Technical Field

[0001] The present application relates to the technical field of radar remote sensing image processing technology, and in particular to a single-polarization synthetic aperture radar change detection method and device. Background Art

[0002] In related technologies, synthetic aperture radar obtains ground object information by actively emitting electromagnetic waves and receiving ground object echoes. It can penetrate clouds and fog and has the advantages of all-day and all-weather operation. Most change monitoring methods based on SAR (Synthetic Aperture Radar) images follow the basic framework of traditional change monitoring, and obtain change results by analyzing dual-phase difference maps. In most difference map analyses, static thresholds are used to segment the global image.

[0003] However, in the related technologies, since SAR images contain a lot of coherent speckle noise, most difference map generation algorithms have limited noise suppression, and the method of using static thresholds to segment the global image is difficult to adapt to the global image. In the process of difference map generation and analysis, the two are relatively independent and not performed end-to-end, resulting in low change monitoring accuracy, which urgently needs to be improved. Summary of the invention

[0004] The present application provides a single-polarization synthetic aperture radar change detection method and device to solve the problems in the related art, such as the limited processing capability of coherent speckle noise in the SAR image monitoring method and the defect that the static threshold segmentation method is difficult to adapt to the global situation, resulting in low accuracy of change monitoring results.

[0005] The first aspect of the present application provides a single-polarization synthetic aperture radar change detection method, comprising the following steps: pre-processing the initial single-polarization SAR images of the previous and next phases in sequence to obtain a final single-polarization SAR image that meets preset processing conditions; calculating a multi-scale feature difference map according to the final single-polarization SAR image, and performing fuzzy C-means clustering based on the multi-scale feature difference map to generate a pseudo label; using the UNet encoding and decoding structure as the basic framework, designing a dual-branch convolutional neural network, using an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer to form a twin convolutional neural network, and establishing a residual connection between the input tensor and the output tensor after network processing to generate a change detection model, and outputting the detection result of the single-polarization synthetic aperture radar using the change detection model trained with the pseudo label.

[0006] The embodiment of the present application can pre-process the initial single-polarization SAR images of the previous and next phases in sequence, so that subsequent analysis and processing can be based on more accurate and reliable data, and generate pseudo labels based on fuzzy C-means clustering based on multi-scale feature difference maps, so as to better handle the uncertainty and ambiguity of the data and improve the accuracy and rationality of classification. The UNet encoding and decoding structure is used as the basic framework, and a twin convolutional neural network is composed of multiple layers in the convolutional neural network. Different layers can extract features at different levels, comprehensively describe the feature information of the single-polarization SAR image, more accurately detect the changed area, and improve the accuracy of change detection.

[0007] Optionally, in one embodiment of the present application, it also includes: adding a spatial-channel attention module to the twin convolutional neural network model to enhance the feature recognition capability of the twin convolutional neural network.

[0008] The embodiment of the present application can enhance the feature recognition capability of the twin convolutional neural network through the space-channel attention module, so that when the network processes single-polarization SAR images, it can pay more attention to the features that have an important impact on the detection results, such as but not limited to the boundaries of the changed area, the changing characteristics of specific objects, etc., thereby improving the recognition accuracy of the changed area.

[0009] Optionally, in one embodiment of the present application, the expression of the spatial attention of the spatial-channel attention module is:

[0010]

[0011] Among them, f C×1×1 is a 1×1 convolution with c output channels, f dc It is a dilated convolutional pyramid;

[0012] The expression of the channel attention of the spatial-channel attention module is:

[0013] M c (F)=MLP(AvgPool(F))=W1(W0AvgPool(F)),

[0014] Among them, W0 and W1 are MLP network parameters, and AvgPool is the global average pooling.

[0015] The spatial attention expression of the embodiment of the present application includes a hole convolution pyramid, which can obtain contextual semantic information feature maps of different scales, so that the network can fuse multi-scale spatial information. The channel attention expression adopts global average pooling, which can obtain the global information of the feature map in each channel, so that the network can understand the feature distribution in the entire channel dimension, thereby improving the recognition accuracy of the changed area.

[0016] Optionally, in one embodiment of the present application, the calculation formula of the multi-scale feature difference map is:

[0017] DI(i,j)=|log(I1(i,j))-log(I2(i,j))|,

[0018] Where i is the row number of the pixel, j is the column number of the pixel, I(i,j) is the pixel value of the i-th row and j-th column in the SAR image, and DI(i,j) is the pixel value of the i-th row and j-th column in the difference image.

[0019] The embodiments of the present application can accurately locate the position of each pixel in the original SAR image and the difference image, accurately quantify the feature difference at each pixel position, and provide an accurate data basis for subsequent analysis and processing.

[0020] Optionally, in one embodiment of the present application, generating a change detection model includes: training an initial change detection model using training samples and a mixed loss function to construct the change detection model, wherein the mixed loss function is:

[0021] Loss=α*DiceLoss+(1-α)*BCELoss,

[0022] Among them, DiceLoss is the Dice coefficient loss, and BCELoss is the binary cross entropy loss.

[0023] The embodiment of the present application can use a mixed loss function including Dice coefficient loss and binary cross entropy loss to train the initial change detection model, guide the model to converge to the optimal parameter value more accurately, and improve the accuracy and reliability of detection.

[0024] The second aspect of the present application provides a single-polarization synthetic aperture radar change detection device, including: a preprocessing module, which is used to preprocess the initial single-polarization SAR images of the previous and next phases in sequence to obtain a final single-polarization SAR image that meets the preset processing conditions; a generation module, which is used to calculate a multi-scale feature difference map according to the final single-polarization SAR image, and perform fuzzy C-means clustering based on the multi-scale feature difference map to generate a pseudo label; an output module, which uses the UNet encoding and decoding structure as the basic framework, designs a dual-branch convolutional neural network, and uses an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer to form a twin convolutional neural network, and establishes a residual connection between the input tensor and the output tensor after network processing to generate a change detection model, and outputs the detection result of the single-polarization synthetic aperture radar using the change detection model trained with the pseudo label.

[0025] The embodiment of the present application can pre-process the initial single-polarization SAR images of the previous and next phases in sequence, so that subsequent analysis and processing can be based on more accurate and reliable data, and generate pseudo labels based on fuzzy C-means clustering based on multi-scale feature difference maps, so as to better handle the uncertainty and ambiguity of the data and improve the accuracy and rationality of classification. The UNet encoding and decoding structure is used as the basic framework, and a twin convolutional neural network is composed of multiple layers in the convolutional neural network. Different layers can extract features at different levels, comprehensively describe the feature information of the single-polarization SAR image, more accurately detect the changed area, and improve the accuracy of change detection.

[0026] Optionally, in one embodiment of the present application, the output module includes: a reinforcement unit for adding a spatial-channel attention module to the twin convolutional neural network model to enhance the feature recognition capability of the twin convolutional neural network.

[0027] The embodiment of the present application can enhance the feature recognition capability of the twin convolutional neural network through the space-channel attention module, so that when the network processes single-polarization SAR images, it can pay more attention to the features that have an important impact on the detection results, such as but not limited to the boundaries of the changed area, the changing characteristics of specific objects, etc., thereby improving the recognition accuracy of the changed area.

[0028] Optionally, in one embodiment of the present application, the expression of the spatial attention of the spatial-channel attention module is:

[0029]

[0030] Among them, f C×1×1 is a 1×1 convolution with c output channels, f dc It is a dilated convolutional pyramid;

[0031] The expression of the channel attention of the spatial-channel attention module is:

[0032] M c (F)=MLP(AvgPool(F))=W1(W0AvgPool(F)),

[0033] Among them, W0 and W1 are MLP network parameters, and AvgPool is the global average pooling.

[0034] The spatial attention expression of the embodiment of the present application includes a hole convolution pyramid, which can obtain contextual semantic information feature maps of different scales, so that the network can fuse multi-scale spatial information. The channel attention expression adopts global average pooling, which can obtain the global information of the feature map in each channel, so that the network can understand the feature distribution in the entire channel dimension, thereby improving the recognition accuracy of the changed area.

[0035] Optionally, in one embodiment of the present application, the calculation formula of the multi-scale feature difference map is:

[0036] DI(i,j)=|log(I1(i,j))-log(I2(i,j))|,

[0037] Where i is the row number of the pixel, j is the column number of the pixel, I(i,j) is the pixel value of the i-th row and j-th column in the SAR image, and DI(i,j) is the pixel value of the i-th row and j-th column in the difference image.

[0038] The embodiments of the present application can accurately locate the position of each pixel in the original SAR image and the difference image, accurately quantify the feature difference at each pixel position, and provide an accurate data basis for subsequent analysis and processing.

[0039] Optionally, in one embodiment of the present application, the output module includes: a construction unit, configured to train an initial change detection model using training samples and a mixed loss function to construct the change detection model, wherein the mixed loss function is:

[0040] Loss=α*DiceLoss+(1-α)*BCELoss,

[0041] Among them, DiceLoss is the Dice coefficient loss, and BCELoss is the binary cross entropy loss.

[0042] The embodiment of the present application can use a mixed loss function including Dice coefficient loss and binary cross entropy loss to train the initial change detection model, guide the model to converge to the optimal parameter value more accurately, and improve the accuracy and reliability of detection.

[0043] The third aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the single-polarization synthetic aperture radar change detection method as described in the above embodiment.

[0044] The fourth aspect embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and when the program is executed by a processor, it implements the above-mentioned single-polarization synthetic aperture radar change detection method.

[0045] The fifth aspect of the present application provides a computer program product, including a computer program, which, when executed, is used to implement the above-mentioned single-polarization synthetic aperture radar change detection method.

[0046] The embodiment of the present application can pre-process the initial single-polarization SAR image to provide accurate and reliable data for the subsequent process, generate pseudo labels by clustering multi-scale feature difference maps, process data uncertainty, and improve classification accuracy. The twin convolutional neural network with UNet as the framework can extract multi-level features and accurately detect changed areas. The feature recognition capability is enhanced through the space-channel attention module, and the key features are focused on to improve accuracy. Its expression integrates multi-scale and global information to further improve accuracy, and can also accurately locate pixel differences to provide a data basis. The model can be trained using a mixed loss function to guide model convergence and improve detection accuracy and reliability.

[0047] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0049] Figure 1 A flow chart of a single-polarization synthetic aperture radar change detection method provided according to an embodiment of the present application;

[0050] Figure 2 A schematic diagram of a twin attention convolutional neural network structure according to a specific embodiment of the present application;

[0051] Figure 3 A schematic diagram of the structure of each submodule of the network according to a specific embodiment of the present application;

[0052] Figure 4 is a schematic diagram of an attention mechanism module according to a specific embodiment of the present application;

[0053] Figure 5 A schematic diagram of the structure of a single-polarization synthetic aperture radar change detection device provided according to an embodiment of the present application;

[0054] Figure 6 The figure is a structural example diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0055] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0056] The following describes the single-polarization synthetic aperture radar change detection method and device of the embodiment of the present application with reference to the accompanying drawings. In view of the related technologies mentioned in the background technology center, due to the limited processing ability of the coherent speckle noise in the SAR image monitoring method and the defect that the static threshold segmentation method is difficult to adapt to the global situation, resulting in the problem that the accuracy of the change monitoring result is not high, the present application provides a single-polarization synthetic aperture radar change detection method, in which the initial single-polarization SAR images of the previous and next phases can be preprocessed in sequence, so that the subsequent analysis and processing can be based on more accurate and reliable data, and fuzzy C-means clustering is performed based on the multi-scale feature difference map to generate pseudo labels, better handle the uncertainty and ambiguity of the data, and improve the accuracy and rationality of the classification. The UNet encoding and decoding structure is used as the basic framework, and multiple layers in the convolutional neural network are used to form a twin convolutional neural network. Different layers can extract features at different levels, comprehensively describe the feature information of the single-polarization SAR image, more accurately detect the change area, and improve the accuracy of change detection. This solves the problems in the related art, such as the limited processing capability of the coherent speckle noise in the SAR image monitoring method and the difficulty of the static threshold segmentation method to adapt to the global situation, which leads to low accuracy of the change monitoring results.

[0057] Specifically, Figure 1 A schematic flow chart of a single-polarization synthetic aperture radar change detection method provided in an embodiment of the present application.

[0058] like Figure 1 As shown, the single-polarization synthetic aperture radar change detection method includes the following steps:

[0059] In step S101, the initial single-polarization SAR images of the preceding and succeeding time phases are preprocessed in sequence to obtain a final single-polarization SAR image that meets a preset processing condition.

[0060] It is understandable that the initial single-polarization SAR image may have various problems, such as but not limited to noise, geometric distortion, inconsistent data format, etc. The purpose of preprocessing is to eliminate or reduce these problems so that the final image can meet the processing conditions, that is, have higher quality and availability, and provide a good data basis for subsequent analysis and processing.

[0061] In some embodiments, for the single-polarization SAR images of the previous and next phases, pre-processing measures such as radiation calibration, terrain correction, filtering and registration are implemented in sequence. As for filtering, the refined Lee filtering method is adopted. Refined Lee filtering has significant advantages. Its calculation amount is relatively small, which will not bring too much calculation burden to the system. At the same time, the filtering effect is relatively good, and it can well maintain the detailed information such as the midpoint, line and edge of the image, and can accurately identify the ground features in the image.

[0062] The embodiment of the present application can eliminate or reduce problems such as noise, geometric distortion, and inconsistent data formats by preprocessing the initial single-polarization SAR image, thereby improving the quality of the image data and making it more suitable for subsequent analysis and processing.

[0063] In step S102, a multi-scale feature difference map is calculated according to the final single-polarization SAR image, and fuzzy C-means clustering is performed based on the multi-scale feature difference map to generate a pseudo label.

[0064] It is understandable that since surface changes may present different characteristics at different scales, calculating multi-scale feature difference maps can comprehensively consider the changes at these different scales and provide richer information for subsequent classification and change detection. Fuzzy C-means clustering is a classification method based on the intrinsic characteristics of the data. This clustering of multi-scale feature difference maps can classify the areas in the image according to the similarity of their features in order to generate pseudo labels. These pseudo labels can be used as a reference for subsequent model training and change detection to help distinguish between changed areas and unchanged areas.

[0065] In the actual implementation process, the single-polarization SAR images of the previous and next phases are calculated using difference or ratio methods to generate a difference map as the input of the network. Since the noise in the SAR image is mainly multiplicative noise, the logarithmic difference method is used to generate the difference map to reduce the noise impact in the SAR image. For the dual-phase SAR images I1 and I2, the calculation formula of the multi-scale feature difference map is:

[0066] DI(i,j)=|log(I1(i,j))-log(I2(i,j))|,

[0067] Where i is the row number of the pixel, j is the column number of the pixel, I(i,j) is the pixel value of the i-th row and j-th column in the SAR image, and DI(i,j) is the pixel value of the i-th row and j-th column in the difference image.

[0068] Furthermore, after the difference map is generated, fuzzy C-means clustering is used to generate pseudo labels. The pseudo labels are generated by performing FCM (Fuzzy C-Means Clustering Algorithm) on the difference map, and the difference map is pre-classified into two categories: change and unchanged. The obtained coarse classification labels are then used to obtain high-confidence labels through the sample selection strategy, and are divided into high-confidence unchanged class Ω0, high-confidence change class Ω1, and uncertain class Ω2. Assuming that the label of the pixel at the spatial position (i, j) of the pre-classification result is Ω, a 5×5 neighborhood window is selected with this position as the center. By voting inside the window, the proportion of the number of pixels in the window with the same label as the center pixel to the total number of pixels in the window is calculated. The calculation method is as follows:

[0069]

[0070] Among them, num(Ω mn =Ω) is the same number as the center pixel label, and p is the proportion of the center pixel label.

[0071] Specifically, by setting a threshold λ, when p ≥ λ, it is considered to be a high-confidence label, otherwise it is a low-confidence label.

[0072] The embodiment of the present application can use the fuzzy C-means clustering method to classify and generate pseudo labels according to the similarity of regional features when clustering multi-scale feature difference maps, which is more in line with the actual feature distribution of image data and improves the accuracy and rationality of classification.

[0073] In step S103, a dual-branch convolutional neural network is designed with the UNet codec structure as the basic framework. The input layer, convolution layer, pooling layer, fully connected layer and output layer are used to form a twin convolutional neural network. A residual connection is established between the input tensor and the output tensor after network processing to generate a change detection model. The change detection model trained with pseudo labels is used to output the detection results of the single-polarization synthetic aperture radar.

[0074] It is understandable that the UNet structure has good performance in the field of image processing. Its encoding and decoding structure can effectively extract the feature information of the image and is suitable for change detection tasks in single-polarization synthetic aperture radar images. A residual connection is established between the input tensor and the output tensor after network processing, which can retain the original input information and avoid information loss in the deep network. At the same time, the residual connection can make the network easier to train, accelerate the convergence process, and enhance the network's ability to express features.

[0075] Specifically, Figure 2 As shown, Figure 2This is a schematic diagram of a twin attention convolutional neural network structure according to a specific embodiment of the present application, wherein Init Block is an initialization module, which is used to initialize various parameters in the network; Downsample is a downsampling module, which is used to reduce the size of the feature map to extract high-level features in the image; Dif_Feature is a feature map difference module, which calculates the two downsampled images to obtain a difference map; Upsamle is an upsampling module, which restores the feature map size reduced during the downsampling process to the original input size; the DCBAM (Dilated Covolution Bottleneck Attention Module) module is a spatial-channel attention module, which is used to enhance the feature recognition ability of the convolutional neural network and improve the detection accuracy of the changed area; OutConv Block is an output module, which converts the feature map learned by the network into the final output.

[0076] Furthermore, if Figure 3 As shown, Figure 3 It is a schematic diagram of the structure of each submodule of the network according to a specific embodiment of the present application. The Init Block (initialization module) includes multiple convolutional layers (Conv(3*3)), batch normalization layers (BN) and activation layers (ReLU). The input data is subjected to feature extraction through the convolutional layer, and the convolution kernel size is 3×3. Then, the data is normalized through the batch normalization layer to make it more stable during the training process. Finally, the data is subjected to nonlinear transformation through the activation layer ReLU to enhance the expression ability of the network; the Downsample (downsampling module) includes a convolutional layer (Conv(3*3)), a batch normalization layer (BN) and an activation layer (ReLU). In addition, There is also a maximum pooling layer (MaxPool). The data first passes through the convolution layer and related processing, and then is downsampled through the maximum pooling layer to reduce the size of the feature map, thereby extracting high-level features in the image; Upsamle (upsampling module) includes a deconvolution layer (ConvTran(2*2)) and a convolution layer (Conv(3*3)), a batch normalization layer (BN) and an activation layer (ReLU). The deconvolution layer is used to restore the feature map size reduced during the downsampling process to the original input size, and then further adjust the feature map through processing such as the convolution layer; OutConv Block (output module) includes a convolution layer and an activation function (Sigmoid). After processing by the previous module, the data is finally converted through this module to convert the feature map learned by the network into the final output. The activation function Sigmoid is used to map the output value to a suitable interval.

[0077] In summary, in order to better realize the change detection of single-polarization synthetic aperture radar, the embodiment of the present application adds a spatial-channel attention module to the twin convolutional neural network model to enhance the feature recognition ability of the twin convolutional neural network, and adopts residual connection. Specifically, the structure of DCBAM can be summarized as: Use residual structure to weight the initial input feature map The final output feature map is obtained The calculation method of F' can be expressed as:

[0078]

[0079] in, is the multiplication of the corresponding position elements of the feature map. The spatial-channel attention M(F) is composed of the spatial attention and channel attention The calculation method is mainly to multiply the two attention feature maps, and then normalize them through the Sigmoid function to get the final attention feature map. s (F), M c The relationship between (F) can be expressed as follows:

[0080]

[0081] Among them, σ() represents the Sigmoid function, and the sizes of the attention feature maps of the two branches are expanded to

[0082] On the one hand, for the spatial attention branch, we first use a 1×1 convolution on the feature map F to reduce the number of channels to C / r (C is the number of channels, r is the scaling factor), then use the dilated convolution pyramid to obtain contextual semantic information feature maps of different scales and concatenate them on the channel dimension, and finally use a 1×1 convolution to reduce the channel dimension to 1 to obtain the spatial attention M. s (F), the expression of spatial attention of the spatial-channel attention module is:

[0083]

[0084] Among them, f C×1×1 is a 1×1 convolution with c output channels, f dc It is a dilated convolutional pyramid.

[0085] On the other hand, for the channel attention branch, the feature map F is first globally average pooled to obtain the channel vector This vector captures the global information of the feature map in each channel. Then the channel vector F cInput into an MLP (Multi-Layer Perception), which consists of two fully connected layers. In order to reduce the parameters, the number of output channels in the first fully connected layer is set to C / r, and then the number of output channels in the second fully connected layer is C. Restore the channel size of the feature map to obtain the channel attention M c (F), the expression of channel attention of the spatial-channel attention module is:

[0086] M c (F)=MLP(AvgPool(F))=W1(W0AvgPool(F)),

[0087] Among them, W0 and W1 are MLP network parameters, and AvgPool is the global average pooling. For a given input feature map After the above calculations, the final feature map weighted by spatial-channel attention can be obtained

[0088] Optionally, in one embodiment of the present application, generating a change detection model includes: training an initial change detection model using training samples and a mixed loss function to construct a change detection model, wherein the mixed loss function is:

[0089] Loss=α*DiceLoss+(1-α)*BCELoss,

[0090] Among them, DiceLoss is the Dice coefficient loss, and BCELoss is the binary cross entropy loss.

[0091] Specifically, DiceLoss is mainly used to measure the overall similarity between two samples, that is, the overall similarity between the predicted image and the label image, and evaluates the prediction results from a macro perspective, while BCELoss compares the similarity between individual pixels and compares the prediction results from a micro perspective. Combining the two can suppress the impact of the imbalance between changed and unchanged pixels. The calculation expressions of the two are as follows:

[0092]

[0093] BCELoss=-[y k **log P k +(1-y k )*log(1-P k )]

[0094] Among them, X and Y are two samples, X∩Y is the number of elements in the intersection between and, |X| is the number of elements in X, |Y| is the number of elements in Y, and y k is the label value of pixel k, p kis the output label probability corresponding to pixel k.

[0095] The embodiments of the present application can effectively extract SAR image features based on the UNet structure, retain the original information through residual connections, accelerate network convergence, enhance feature expression capabilities, and use the space-channel attention module to improve feature recognition accuracy. It can also suppress category imbalance by integrating macro and micro similarities through a hybrid loss function, guide the model to converge accurately, thereby improving monitoring accuracy and reliability and reducing training time.

[0096] According to the single-polarization synthetic aperture radar change detection method proposed in the embodiment of the present application, the initial single-polarization SAR images of the previous and subsequent phases can be preprocessed in sequence, so that subsequent analysis and processing can be based on more accurate and reliable data. Fuzzy C-means clustering is performed based on multi-scale feature difference maps to generate pseudo labels, better handle the uncertainty and ambiguity of the data, and improve the accuracy and rationality of classification. The UNet encoding and decoding structure is used as the basic framework, and multiple layers in the convolutional neural network are used to form a twin convolutional neural network. Different layers can extract features at different levels, comprehensively describe the feature information of the single-polarization SAR image, more accurately detect the change area, and improve the accuracy of change detection.

[0097] Next, the single-polarization synthetic aperture radar change detection device proposed according to the embodiment of the present application is described with reference to the accompanying drawings.

[0098] Figure 5 It is a block diagram of a single-polarization synthetic aperture radar change detection device according to an embodiment of the present application.

[0099] like Figure 5 As shown, the single-polarization synthetic aperture radar change detection device 10 includes: a preprocessing module 100, a generating module 200 and an output module 300.

[0100] Specifically, the preprocessing module 100 is used to preprocess the initial single-polarization SAR images of the preceding and succeeding time phases in sequence to obtain a final single-polarization SAR image that meets a preset processing condition.

[0101] The generating module 200 is used to calculate a multi-scale feature difference map according to the final single-polarization SAR image, and perform fuzzy C-means clustering based on the multi-scale feature difference map to generate a pseudo label.

[0102] The output module 300 is used to design a dual-branch convolutional neural network with the UNet encoding and decoding structure as the basic framework, use the input layer, convolution layer, pooling layer, fully connected layer and output layer to form a twin convolutional neural network, and establish a residual connection between the input tensor and the output tensor after network processing to generate a change detection model, and use the change detection model obtained by pseudo-label to output the detection result of the single-polarization synthetic aperture radar.

[0103] Optionally, in one embodiment of the present application, the output module 300 includes: a reinforcement unit for adding a spatial-channel attention module to the twin convolutional neural network model for enhancing the feature recognition capability of the twin convolutional neural network.

[0104] Optionally, in one embodiment of the present application, the expression of spatial attention of the spatial-channel attention module is:

[0105]

[0106] Among them, f C×1×1 is a 1×1 convolution with c output channels, f dc It is a dilated convolutional pyramid.

[0107] The expression of channel attention of the spatial-channel attention module is:

[0108] M c (F)=MLP(AvgPool(F))=W1(W0AvgPool(F)),

[0109] Among them, W0 and W1 are MLP network parameters, and AvgPool is the global average pooling.

[0110] Optionally, in one embodiment of the present application, the calculation formula of the multi-scale feature difference map is:

[0111] DI(i,j)=|log(I1(i,j))-log(I2(i,j))|,

[0112] Where i is the row number of the pixel, j is the column number of the pixel, I(i,j) is the pixel value of the i-th row and j-th column in the SAR image, and DI(i,j) is the pixel value of the i-th row and j-th column in the difference image.

[0113] Optionally, in one embodiment of the present application, the output module 300 includes: a construction unit, configured to train an initial change detection model using training samples and a hybrid loss function to construct a change detection model, wherein the hybrid loss function is:

[0114] Loss=α*DiceLoss+(1-α)*BCELoss,

[0115] Among them, DiceLoss is the Dice coefficient loss, and BCELoss is the binary cross entropy loss.

[0116] It should be noted that the above explanation of the embodiment of the single-polarization synthetic aperture radar change detection method is also applicable to the single-polarization synthetic aperture radar change detection device of this embodiment, which will not be repeated here.

[0117] According to the single-polarization synthetic aperture radar change detection device proposed in the embodiment of the present application, the initial single-polarization SAR images of the previous and next phases can be preprocessed in sequence, so that subsequent analysis and processing can be based on more accurate and reliable data. Fuzzy C-means clustering is performed based on multi-scale feature difference maps to generate pseudo labels, better handle the uncertainty and ambiguity of the data, and improve the accuracy and rationality of classification. The UNet encoding and decoding structure is used as the basic framework, and multiple layers in the convolutional neural network are used to form a twin convolutional neural network. Different layers can extract features at different levels, comprehensively describe the feature information of the single-polarization SAR image, more accurately detect the change area, and improve the accuracy of change detection.

[0118] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0119] A memory 601 , a processor 602 , and a computer program stored in the memory 601 and executable on the processor 602 .

[0120] When the processor 602 executes the program, the single-polarization synthetic aperture radar change detection method provided in the above embodiment is implemented.

[0121] Furthermore, the electronic device further comprises:

[0122] The communication interface 603 is used for communication between the memory 601 and the processor 602 .

[0123] The memory 601 is used to store computer programs that can be executed on the processor 602 .

[0124] The memory 601 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0125] If the memory 601, the processor 602 and the communication interface 603 are implemented independently, the communication interface 603, the memory 601 and the processor 602 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0126] Optionally, in a specific implementation, if the memory 601, the processor 602 and the communication interface 603 are integrated on a chip, the memory 601, the processor 602 and the communication interface 603 can communicate with each other through an internal interface.

[0127] The processor 602 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0128] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above single-polarization synthetic aperture radar change detection method.

[0129] An embodiment of the present application further provides a computer program product, including a computer program, which, when executed, is used to implement the above single-polarization synthetic aperture radar change detection method.

[0130] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0131] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0132] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0133] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.

[0134] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0135] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0136] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0137] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for detecting changes in a single-polarization synthetic aperture radar, characterized in that: The following steps are involved: Preprocessing the initial single-polarization SAR images of the previous and next phases in turn to obtain a final single-polarization SAR image that meets the preset processing conditions; Calculating a multi-scale feature difference map according to the final single-polarization SAR image, and performing fuzzy C-means clustering based on the multi-scale feature difference map to generate a pseudo label; Taking the UNet encoding and decoding structure as the basic framework, a dual-branch convolutional neural network is designed. The input layer, convolutional layer, pooling layer, fully connected layer and output layer are used to form a twin convolutional neural network. A residual connection is established between the input tensor and the output tensor after network processing to generate a change detection model. The change detection model trained with the pseudo-labels is used to output the detection results of the single-polarization synthetic aperture radar.

2. The method according to claim 1, characterized in that Also includes: A spatial-channel attention module is added to the twin convolutional neural network model to enhance the feature recognition capability of the twin convolutional neural network.

3. The method according to claim 2, characterized in that in, The expression of the spatial attention of the spatial-channel attention module is: Among them, f C×1×1 is a 1×1 convolution with C output channels, f dc It is a dilated convolutional pyramid; The expression of the channel attention of the spatial-channel attention module is: M c (F)=MLP(AvgPool(F))=W1(W0AvgPool(F)), Among them, W0 and W1 are MLP network parameters, and AvgPool is the global average pooling.

4. The method according to claim 1, characterized in that: The calculation formula of the multi-scale feature difference map is: DI(i,j)=|log(I1(i,j))-log(I2(i,j))|, Where i is the row number of the pixel, j is the column number of the pixel, I(i,j) is the pixel value of the i-th row and j-th column in the SAR image, and DI(i,k) is the pixel value of the i-th row and j-th column in the difference image.

5. The method according to claim 1, characterized in that The generating of the change detection model comprises: The initial change detection model is trained using the training samples and the mixed loss function to construct the change detection model, wherein the mixed loss function is: Loss=α*DiceLoss+(1-α)*BCELoss, Among them, DiceLoss is the Dice coefficient loss, and BCELoss is the binary cross entropy loss.

6. A single-polarization synthetic aperture radar change detection device, characterized in that: include: A preprocessing module is used to preprocess the initial single-polarization SAR images of the previous and next phases in sequence to obtain a final single-polarization SAR image that meets the preset processing conditions; A generating module, configured to calculate a multi-scale feature difference map according to the final single-polarization SAR image, and perform fuzzy C-means clustering based on the multi-scale feature difference map to generate a pseudo label; The detection module uses the UNet encoder-decoder structure as the basic framework, designs a dual-branch convolutional neural network, uses the input layer, convolution layer, pooling layer, fully connected layer and output layer to form a twin convolutional neural network, and establishes a residual connection between the input tensor and the output tensor after network processing to generate a change detection model. The change detection model trained with pseudo labels outputs the detection results of the single-polarization synthetic aperture radar.

7. The device according to claim 6, characterized in that The output module comprises: The strengthening unit is used to add a spatial-channel attention module to the twin convolutional neural network model to strengthen the feature recognition ability of the twin convolutional neural network.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the single-polarization synthetic aperture radar change detection method according to any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the single-polarization synthetic aperture radar change detection method as described in any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that The computer program is executed to implement the single-polarization synthetic aperture radar change detection method according to any one of claims 1 to 5.

Images