Anti-noise interference polarimetric SAR image change detection method and system

Abstract

The invention discloses an anti-noise interference polarized SAR image change detection method and system. The change detection method comprises the following steps: acquiring a polarized SAR image; inputting the acquired polarized SAR image into the trained denoising and change detection model, and outputting a denoising result and a change detection prediction result; the training of the denoising and change detection model comprises the following steps: constructing a data set by adopting a covariance matrix component and a polarization decomposition component of a polarized SAR image; constructing a denoising network and a change detection network, wherein the change detection network is a transformer network in which space attention and channel attention are embedded; training a denoising network through the data set, obtaining denoising results of the polarized SAR images at the same place and at different times, inputting the denoising results into a change detection network for training, converting deep features extracted by the denoising network into attention weights, and inputting the attention weights into the change detection network to re-calibrate the recognition capability; the method has the advantages of being accurate in detection and good in denoising effect.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image processing, and in particular relates to a polarimetric SAR image change detection method and system under strong noise interference. Background Art

[0002] Change detection is the process of identifying the difference in the state of an object or phenomenon by observing it at different times. It is one of the main problems in earth observation and has been widely studied in recent decades. Synthetic aperture radar (SAR) is widely used in various fields because its imaging process is not affected by sunlight conditions, cloud cover and atmospheric conditions. Multi-polarization SAR images can provide richer scattering information than single-polarization SAR images to identify the differences in land use and land cover in a specific area over a period of time. This is very important in various applications such as urban planning, environmental monitoring, agricultural surveys, disaster assessment and map revision.

[0003] There are three main categories of existing SAR image change detection methods. The first category is traditional clustering / classification-based methods, which usually use traditional clustering or classification algorithms to divide data into changed areas and unchanged areas, but have high computational complexity and poor accuracy. The second category is deep learning methods based on convolutional neural networks (CNNs), which use CNNs to automatically learn feature representations from data, thereby capturing complex patterns and spatial dependencies between pixels. Compared with traditional methods, this method has been shown to have better performance, but is limited by the receptive field. The third category is Transformer-based deep learning methods, which use self-attention mechanisms to weight different areas of the input image according to the relevance of the input image to the task, or capture long-distance dependencies between pixels and generate feature representations. Some scholars have tried to combine CNN and Transformer structures to take full advantage of the advantages of both. These methods have shown excellent performance compared to previous methods and represent the latest progress in remote sensing image change detection.

[0004] However, these existing methods still face some problems. First, due to the influence of its imaging mechanism, SAR images have inherent coherent speckle noise, which greatly interferes with the accuracy of change detection. Second, the current SAR image change detection methods do not make full use of the rich scattering information in SAR image data, resulting in a decrease in detection accuracy. Therefore, it is very necessary to study how to make full use of the scattering information in SAR images and the change detection method that is resistant to noise interference. Summary of the invention

[0005] The present invention provides a polarimetric SAR image change detection method and system resistant to noise interference. Aiming at the problem that the existing SAR image change detection method does not make sufficient use of polarimetric scattering information, a data set consisting of covariance matrix components and polarimetric decomposition components of polarimetric SAR images is used to replace the common intensity image data set, and a transformer change denoising and change detection model combining spatial attention, channel attention and cross attention is constructed for training. Aiming at the problem that the inherent coherent speckle noise in SAR images has a negative impact on the change detection accuracy, a denoising network is combined to first perform preliminary denoising on the noisy image to obtain a more accurate change detection result. The scheme has the advantages of accurate detection and good denoising effect.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a method for detecting changes in polarimetric SAR images with resistance to noise interference, comprising the following steps:

[0007] Acquire polarimetric SAR images;

[0008] The acquired polarimetric SAR image is input into the trained denoising and change detection model, and the denoising result and the change detection prediction result are output; wherein the training of the denoising and change detection model includes:

[0009] The data set is constructed using the covariance matrix components and polarization decomposition components of polarimetric SAR images;

[0010] Construct a denoising network and a change detection network, wherein the denoising network is a residual convolutional network with embedded channel attention and cross attention, and the change detection network is a transformer network with embedded spatial attention and channel attention;

[0011] The denoising network is trained using the data set to obtain denoising results of polarimetric SAR images of the same location at different times, and the denoising results are input into the change detection network for training. At the same time, the deep features extracted by the denoising network are converted into attention weights and input into the change detection network to recalibrate the recognition capability.

[0012] In the above scheme, during the test phase, the trained denoising and change detection model denoises and identifies the change area of ​​the polarimetric SAR image under strong noise conditions, and obtains the corresponding denoising results and change detection prediction results. During the model training process, the model is trained by a data set constructed by covariance matrix components and polarization decomposition components, which can make full use of the polarimetric scattering information of the image; the residual convolutional network constructed is used for polarimetric SAR self-supervised denoising, and the denoising results are used to train the change monitoring network, and the denoising results and the deep features extracted by the denoising network are input into the change detection network, where the denoising results at different times are used as the training data of the change detection network, and the deep features extracted by the denoising network are used to effectively extract and utilize the image information and improve the model recognition accuracy.

[0013] Furthermore, the step of constructing the data set includes:

[0014] For two polarimetric SAR images at different times in the same scene, their covariance matrix components and polarimetric decomposition components are calculated respectively;

[0015] The covariance matrix components and polarization decomposition components of each temporal SAR image are concatenated along the channel dimension to construct a dataset.

[0016] Furthermore, the step of constructing the denoising network includes:

[0017] Constructing the residual convolutional network, wherein the residual convolutional network extracts features hierarchically by a plurality of continuous dense residual modules and is connected together by residual connections;

[0018] Construct channel attention module 1 to re-weight the feature layer of the denoising network in the channel direction;

[0019] A cross-attention module is constructed to fuse the feature maps of polarimetric SAR images before and after denoising.

[0020] Through continuous dense residual modules, the number of network layers is increased and deeper features can be extracted; the channel attention module is used to enhance the ability to extract polarized information, and the cross attention module uses the respective advantages of the features before and after denoising to complement information and achieve a balance between smooth information and texture information.

[0021] Furthermore, the step of constructing the denoising network includes constructing the following denoising loss function and calculating a preliminary denoising result:

[0022] Denoising loss function:

[0023] (1)

[0024] in represents the mean square error, represent Polarimetric SAR data denoising results for time, Represents the polarimetric SAR data of the same area at another time. Mask is the mask calculated based on the change detection results.

[0025] The mask guides the denoising network to pay more attention to the areas with less changes.

[0026] Furthermore, the step of constructing the change detection network includes:

[0027] Construct a feature extraction module to extract features from polarimetric SAR images of the same location at different times based on the ResNet network;

[0028] A spatial attention module and a channel attention module 2 are constructed to re-weight the channel dimension and spatial dimension of the features extracted from each temporal polarimetric SAR image.

[0029] Furthermore, the step of constructing the change detection network also includes:

[0030] The recalibrated polarimetric SAR image feature map at each time point is converted into a semantic token through a change detection network.

[0031] The Transformer encoder structure is used to enhance the global information of semantic tokens at different times.

[0032] The enhanced semantic Token is reconstructed through the twin Transformer decoder structure to obtain the reconstructed feature map of the corresponding time.

[0033] Furthermore, the step of constructing the change detection network also includes:

[0034] Construct the change detection loss function shown below to calculate the differential features and output the change detection prediction results:

[0035] (2)

[0036] in represents the cross entropy loss, is the result predicted by the model at position (h, w), is the label at position (h, w), h and w are the position coordinates, and H and W are the upper limits of the position coordinates.

[0037] Furthermore, the network training is constrained by the following two-branch joint loss function:

[0038] (3)

[0039] in and are the loss functions of the change detection network and the denoising network, respectively. and are the weights of change detection loss and denoising loss respectively.

[0040] A polarimetric SAR image change detection system resistant to noise interference, comprising:

[0041] Data acquisition module, used to acquire polarimetric SAR images;

[0042] The change detection module is used to input the acquired polarimetric SAR image into the trained denoising and change detection model, and output the denoising result and the change detection prediction result; wherein the training of the denoising and change detection model includes:

[0043] The data set is constructed using the covariance matrix components and polarization decomposition components of polarimetric SAR images;

[0044] Construct a denoising network and a change detection network, wherein the denoising network is a residual convolutional network with embedded channel attention and cross attention, and the change detection network is a transformer network with embedded spatial attention and channel attention;

[0045] The denoising network is trained using the data set to obtain denoising results of polarimetric SAR images of the same location at different times, and the denoising results are input into the change detection network for training. At the same time, the deep features extracted by the denoising network are converted into attention weights and input into the change detection network to recalibrate the recognition capability.

[0046] A noise-resistant polarimetric SAR image change detection device comprises a memory and a processor, wherein the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the steps of the above method.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] (1) Using the covariance matrix and polarization decomposition components as training data can make full use of the rich polarization scattering information in polarimetric SAR images, thereby more accurately identifying the change area;

[0049] (2) By embedding spatial attention and channel attention in the transformer structure, a change detection network is constructed that can effectively extract and utilize polarization scattering information;

[0050] (3) To address the problem that high-intensity coherent speckle noise in polarimetric SAR may cause the change detection network to be unable to accurately distinguish the change area, a denoising network system combining channel attention, cross attention and residual convolutional network is designed to denoise noisy images at different times. At the same time, the extracted deep denoising features are converted into attention weights to guide the change detection network to effectively identify the contours and boundaries of objects. By combining the denoising network for auxiliary training, change detection that is resistant to noise interference can be achieved. In addition, a two-branch joint loss function is used for constrained training to achieve collaborative optimization between the change detection and denoising networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a principle diagram of a method for detecting changes in polarized SAR images with noise immunity provided by an embodiment of the present invention;

[0052] Figure 2 It is a schematic diagram of a network framework of a method for detecting changes in polarized SAR images with noise immunity provided by an embodiment of the present invention;

[0053] Figure 3 It is a training and testing flow chart of a noise-resistant polarimetric SAR image change detection method provided by an embodiment of the present invention;

[0054] Figure 4 This is the structure of the transformer encoder and decoder provided in the embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention are arbitrarily combined with each other to form a new technical solution. This combination is not restricted by the sequence of steps and / or the structural composition mode, but must be based on the ability of ordinary technicians in this field to achieve. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that this combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0056] like Figure 1 , Figure 3 As shown, a method for detecting changes in polarimetric SAR images with resistance to noise interference comprises the following steps:

[0057] Acquire polarimetric SAR images;

[0058] The acquired polarimetric SAR image is input into the trained denoising and change detection model, and the denoising result and the change detection prediction result are output; wherein the training of the denoising and change detection model includes:

[0059] The data set is constructed using the covariance matrix components and polarization decomposition components of polarimetric SAR images;

[0060] Construct a denoising network and a change detection network, wherein the denoising network is a residual convolutional network with embedded channel attention and cross attention, and the change detection network is a transformer network with embedded spatial attention and channel attention;

[0061] The denoising network is trained using the data set to obtain denoising results of polarimetric SAR images of the same location at different times, and the denoising results are input into the change detection network for training. Meanwhile, the deep features extracted by the denoising network are converted into attention weights and input into the change detection network to recalibrate the recognition capability.

[0062] In the above scheme, during the test phase, the trained denoising and change detection model performs denoising and change area identification on the polarimetric SAR image under strong noise conditions, and obtains the corresponding denoising results and change detection prediction results. During the model training process, the model is trained by a data set constructed by covariance matrix components and polarization decomposition components, which can make full use of the polarimetric scattering information of the image; in the step of constructing the model denoising network and change detection network framework, the residual convolution network constructed is used for polarimetric SAR self-supervised denoising, and the denoising results are used to train the change monitoring network, and the denoising results and the deep features extracted by the denoising network are input into the change detection network, where the denoising results at different times are used as the training data of the change detection network, and the deep features extracted by the denoising network are used to effectively extract and utilize the image information and improve the model recognition accuracy.

[0063] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0064] The image is preprocessed by radiation correction, terrain correction, multi-view processing, etc. to obtain the corrected image. The covariance matrix components and polarization decomposition components of the polarization SAR images at two different times in the same scene are calculated to construct a data set. Currently, multi-polarization SAR includes dual polarization and full polarization. Full polarization SAR data can ultimately be expressed as nine covariance matrix components and three polarization decomposition components. Dual polarization SAR consists of four covariance matrix components and three polarization decomposition components. According to the different polarization modes, the corresponding SAR image is selected to calculate the two parameters.

[0065] For two polarimetric SAR images at different times in the same scene ( , ), the scattering matrix is ​​usually used to represent the information of each element of the SAR image.

[0066] A. Assuming it is a fully polarized SAR image, the scattering matrix It can be expressed as:

[0067] (4)

[0068] In the formula, express Launch and Received scattering elements. represents a horizontally polarized wave, Represents a vertically polarized wave.

[0069] In practical applications, the two-dimensional scattering matrix cannot effectively characterize all the characteristics of the target, so the second-order statistics of the scattering matrix are introduced to analyze its electromagnetic scattering characteristics. Assuming that the scattering matrix satisfies the reciprocity theorem, when by When the basis is expressed, When , its covariance matrix It can be expressed as:

[0070] (5)

[0071] in represents the set average operator and * represents the complex conjugate operation. Represents the modulus of a complex signal.

[0072] Extract the real and imaginary parts of the upper triangular matrix of the covariance matrix to obtain the SAR covariance value matrix As the components of the input data. The formula of the covariance value matrix is ​​as follows:

[0073] (6)

[0074] The subscripts indicate their positions in the covariance matrix. represents the real value, Represents the imaginary part value.

[0075] For further analysis of the data, the Yamaguchi three-component (YAM3) polarimetric decomposition was used to extract the scattered components that represent the characteristics of various objects, which were then added as an additional part of the input data. The YAM3 polarimetric decomposition has three components: , ,and . Corresponds to single scattering, representing surfaces, water, and flat terrain. Corresponds to double scattering, commonly seen in the dihedral angles of buildings. Corresponding volume scattering, mainly represents the vegetation in the scene.

[0076] The constructed full polarimetric SAR dataset contains two time periods, and the data at each time period consists of 9 covariance matrix components and 3 polarization decomposition components.

[0077] B. Assuming it is a dual-polarization SAR image, the scattering matrix should be adjusted accordingly according to its polarization mode. For example, for VV and VH polarizations, the scattering matrix Adjusted to:

[0078] (7)

[0079] Its covariance matrix The corresponding adjustments are:

[0080] (8)

[0081] The covariance value matrix is:

[0082] (9)

[0083] To further analyze the data, the Raney polarization decomposition method is used to extract the scattered components representing the characteristics of various objects, which are then added as an additional part of the input data. Raney polarization decomposition has three components: , ,and . Corresponds to single scattering, representing surfaces, water, and flat terrain. Corresponds to double scattering, commonly seen in the dihedral angles of buildings. Corresponding volume scattering, mainly represents the vegetation in the scene.

[0084] The constructed dual-polarization SAR dataset contains two time periods, and the data at each time period consists of 4 covariance matrix components and 3 polarization decomposition components.

[0085] The covariance matrix components and polarization decomposition components of the polarimetric SAR images at each time are concatenated along the channel dimension to construct a dataset.

[0086] Construct a denoising network and a change detection network, and construct a residual convolutional network with embedded channel attention and cross attention for polarimetric SAR self-supervised denoising. Figure 2 As shown in the denoising network in , a polarimetric SAR image of the same location at one time is used as the noise image, and a polarimetric SAR image of another time is used as the reference image for self-supervised denoising.

[0087] Construct a residual convolutional network framework. The residual convolutional network first performs preliminary feature extraction by two consecutive 3×3 convolutional layers, then extracts residual features hierarchically by three consecutive dense residual modules, and connects the hierarchical features together through residual connections to ensure that deeper features can be extracted. Then, the number of channels is adjusted to be consistent with the preliminary extracted features through 1×1 convolution, and deep residual features are obtained through 3×3 convolution. The deep features are added to the preliminary features to obtain enhanced features, and finally the final result is output through a 3×3 convolutional layer.

[0088] Construct channel attention module 1. Use channel attention after each dense residual module in the residual convolutional network framework to recalibrate the extracted polarimetric SAR features in the polarimetric channel dimension, thereby effectively maintaining the polarimetric information. This module can be defined as:

[0089] (10)

[0090] in, represents the feature layer that has been weighted by the channel attention module. represents the channel attention input feature layer, represents the average pooling operation, and Represent the ReLU activation function and the Sigmoid activation function respectively. , and and Correspondingly represent the weight term and the bias term, represents element-wise multiplication. Channel attention is achieved through Calculate the channel attention feature map and multiply it with the input feature to recalibrate it in the channel dimension.

[0091] Construct a cross-attention module. This module is used to fuse the feature maps of polarimetric SAR images before and after denoising. Usually, the features before denoising contain more spatial information, and the features after denoising are smoother. The cross-attention module uses the rich spatial information before denoising to guide the generation of features after denoising, and uses the polarimetric information after denoising to know the generation before denoising, thereby achieving a balance between smooth features and texture features. This module can be defined as:

[0092] (11)

[0093] in and Represent the features before and after denoising respectively. and The feature maps before and after denoising are obtained respectively. , The weights of the features before and after denoising are obtained respectively.

[0094] The cross attention module uses the denoised feature weights to recalibrate the feature map before denoising to enhance the polarization information of the features before denoising. , use the feature weights before denoising to recalibrate the denoised feature map to enhance the spatial information of the denoised features , the two enhanced features are fused and obtained .

[0095] Construct the denoising loss function shown below and calculate the preliminary denoising results.

[0096] Denoising loss function:

[0097] (12)

[0098] in represents the mean square error, represent Polarimetric SAR data denoising results for time, Represents the polarimetric SAR data of the same area at another time. Mask is a mask calculated based on the change detection results, which is used to guide the denoising network to pay more attention to areas with smaller changes.

[0099] The denoising network is trained using the dataset to obtain the denoising results of polarimetric SAR images at the same location at different times, and a transformer network with embedded spatial attention and channel attention is constructed for change detection. Figure 2 As shown in the change detection network in , its characteristics are: taking two denoised images of the same place at different times as input data and the change detection label between the two as reference, the change detection network is trained. The change detection network consists of a feature extraction module, a channel and spatial attention module, a transformer encoder, and two transformer decoders.

[0100] Construct a feature extraction module. The feature extraction module is an improved ResNet18 network. The ResNet18 network has 5 stages. The stride of the last 3 stages is replaced with 1, and the feature dimension is reduced by point-by-point convolution (output channel is 32). Features are extracted from polarimetric SAR images of the same location at different times.

[0101] Construct the second spatial and channel attention module. The spatial attention module is used to recalibrate the extracted polarimetric SAR features in the polarimetric channel dimension, thereby effectively maintaining the polarimetric information. This module can be defined as:

[0102] (13)

[0103] in, represents the feature layer re-weighted by the spatial attention module, Represents the spatial attention input feature layer.

[0104] Spatial attention through The spatial attention weight map is calculated and multiplied with the input features to recalibrate the spatial dimension.

[0105] The channel attention module 2 in the change detection network has the same structure and function as the channel attention module 1 in the denoising network.

[0106] The deep features extracted by the denoising network are converted into attention weights through 1×1 convolution and Sigmoid activation function and input into the change detection network to guide the recalibration of the features after spatial and channel attention and enhance the change detection network's ability to resist noise interference.

[0107] The 3×3 convolution and softmax function are used to convert the feature maps of each time after multiple calibration into corresponding semantic tokens and concatenate them.

[0108] The Transformer encoder structure is used to enhance the global information of the concatenated semantic tokens. The structure of the encoder is as follows: Figure 4 As shown. It consists of a multi-head self-attention and a multi-layer perceptron. The multi-head self-attention consists of multiple attention heads. Each attention head calculates the query (Q), key (K), and value (V) based on the intermediate token. Then Q and K are used to calculate the attention score, which is then used to weight V. The multi-head self-attention connects the outputs of all attention heads and performs a linear transformation. This process can be expressed as follows:

[0109]

[0110]

[0111] (14)

[0112] in represents multi-head self-attention, Represents the concatenation of the Token set input from the previous layer, is the transpose operation, It is a calculation structure The process of attention head; , , and is the linear projection matrix; represents the number of attention heads, The parameters are The channel dimension.

[0113] Afterwards, it is processed using a multi-layer perceptron to generate new tokens. MLP mainly performs nonlinear transformation on the input sequence to enhance its expressiveness.

[0114] Construct a twin Transformer decoder structure to reconstruct the enhanced semantic Token. The structure of the decoder is as follows: Figure 4 As shown in Figure 1. It consists of a multi-head cross attention and a multi-layer perceptron. The multi-head cross attention also consists of multiple attention heads. Each attention head combines the new token with the input features of the corresponding time to calculate the Q, K, and V matrices. The multi-head cross attention also connects the outputs of all attention heads and performs a linear transformation.

[0115]

[0116] (15)

[0117] in It is multi-headed cross-attention. and , They are the decoded and reconstructed features and the enhanced Token at two different times.

[0118] Afterwards, a multi-layer perceptron is used to reconstruct new data containing global context information at the corresponding time.

[0119] Construct the change detection loss function shown below to calculate the differential features and output the change detection prediction results.

[0120] Change detection loss function:

[0121] (16)

[0122] in represents the cross entropy loss, is the result predicted by the model at position (h,w), is the label at position (h,w), H and W are both constants, representing the upper limit of the position coordinates respectively.

[0123] like Figure 3 As shown in the figure. In the training phase, the constructed training data set is first used to train the constructed denoising network to obtain the denoising results of polarimetric SAR images of the same location at different times. The denoising results of polarimetric SAR images of the same location at different times are then input into the constructed change detection network for training. At the same time, the deep features extracted by the denoising network are converted into attention weights to guide the change detection network to effectively identify the contours and boundaries of objects.

[0124] In the training phase, we first use the constructed training data set to train the constructed denoising network. , The temporal polarimetric SAR image is a noise image. , The polarimetric SAR image of time is used as the reference image for denoising to obtain , Denoising results of temporal polarimetric SAR images , . Then , The constructed change detection network is trained by inputting it, and the deep features extracted by the denoising network are converted into attention weights to guide the change detection network to effectively identify the contours and boundaries of objects. , The polarization covariance matrix Input is the change detection results generated by the PolSARpro software developed by ESA.

[0125] When training the denoising network, we use , The temporal polarimetric SAR image is a noise image. , The polarimetric SAR image of time is used as the reference image for denoising to obtain , Denoising results of temporal polarimetric SAR images , In the first epoch of training, the mask for calculating the denoising loss is 1. Starting from the second epoch, the change detection result of the previous epoch is used as the mask for calculating the denoising loss.

[0126] Then , The input change detection network is trained, and the deep features extracted by the denoising network are converted into attention weights. , The polarization covariance matrix The input is the change detection results generated by the PolSARpro software developed by ESA. The number of layers of the Transformer encoder is set to 1, and the number of layers of the Transformer decoder is set to 8. The number of heads h of MSA and MA is set to 8, and the channel dimension of each head is set to 8.

[0127] The training of the network is constrained by the following two-branch joint loss function.

[0128] Total loss function:

[0129] (17)

[0130] in and are the loss functions of the change detection network and the denoising network, respectively. and are the weights of the change detection loss function and the denoising loss function, respectively.

[0131] , .

[0132] In the testing phase, the trained denoising and change detection networks are used to denoise the polarimetric SAR images and identify change areas, and the corresponding denoising results and change detection prediction results are obtained.

[0133] The embodiment of the present application also provides a polarimetric SAR image change detection system that is resistant to noise interference, including:

[0134] The first main module is used to construct a data set using covariance matrix components and polarization decomposition components of polarimetric SAR images;

[0135] The second main module is used to perform training through the data set to obtain denoising results of polarimetric SAR images at the same location and at different times, and to extract deep features of polarimetric SAR images;

[0136] The third main module is used to train with the denoising results and calibrate the recognition capability according to the attention weights converted from the deep features.

[0137] An embodiment of the present application also provides a noise-resistant polarization SAR image change detection device, including a memory and a processor, wherein the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the steps of the above method.

[0138] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A noise-resistant polarimetric SAR image change detection method, characterized in that: The steps include: Acquire polarimetric SAR images; The acquired polarimetric SAR image is input into the trained denoising and change detection model, and the denoising result and the change detection prediction result are output; wherein the training of the denoising and change detection model includes: The data set is constructed using the covariance matrix components and polarization decomposition components of polarimetric SAR images; Construct a denoising network and a change detection network, wherein the denoising network is a residual convolutional network with embedded channel attention and cross attention, and the change detection network is a transformer network with embedded spatial attention and channel attention; The denoising network is trained using the data set to obtain denoising results of polarimetric SAR images of the same location at different times, and the denoising results are input into the change detection network for training. Meanwhile, the deep features extracted by the denoising network are converted into attention weights and input into the change detection network to recalibrate the recognition capability.

2. The noise-resistant polarimetric SAR image change detection method according to claim 1, characterized in that: The steps of constructing the dataset include: For two polarimetric SAR images at different times in the same scene, their covariance matrix components and polarimetric decomposition components are calculated respectively; The covariance matrix components and polarization decomposition components of each temporal SAR image are concatenated along the channel dimension to construct a dataset.

3. The noise-resistant polarimetric SAR image change detection method according to claim 1, characterized in that: The steps of constructing the denoising network include: Constructing the residual convolutional network, wherein the residual convolutional network extracts features hierarchically by a plurality of continuous dense residual modules and is connected together by residual connections; Construct channel attention module 1 to re-weight the feature layer of the denoising network in the channel direction; A cross-attention module is constructed to fuse the feature maps of polarimetric SAR images before and after denoising.

4. The noise-resistant polarimetric SAR image change detection method according to claim 3 is characterized in that: The steps of constructing the denoising network include constructing the following denoising loss function and calculating the preliminary denoising result: in represents the mean square error, represent Polarimetric SAR data denoising results for time, Represents the polarimetric SAR data of the same area at another time. Mask is the mask calculated based on the change detection results.

5. The noise-resistant polarimetric SAR image change detection method according to claim 1, characterized in that: The steps of constructing the change detection network include: Construct a feature extraction module to extract features from polarimetric SAR images of the same location at different times based on the ResNet network; A spatial attention module and a channel attention module 2 are constructed to re-weight the channel dimension and spatial dimension of the features extracted from each temporal polarimetric SAR image.

6. The noise-resistant polarimetric SAR image change detection method according to claim 5, characterized in that: The step of constructing the change detection network also includes: The recalibrated polarimetric SAR image feature map at each time point is converted into a semantic token through a change detection network. The Transformer encoder structure is used to enhance the global information of semantic tokens at different times. The enhanced semantic Token is reconstructed through the twin Transformer decoder structure to obtain the reconstructed feature map of the corresponding time.

7. The noise-resistant polarimetric SAR image change detection method according to claim 6, characterized in that: The step of constructing the change detection network also includes: Construct the change detection loss function shown below to calculate the differential features and output the change detection prediction results: in represents the cross entropy loss, is the result predicted by the model at position (h, w), is the label at position (h,w), h and w represent the position of the pixel in the feature map, and H and W represent the size of the feature map.

8. The noise-resistant polarimetric SAR image change detection method according to claim 1, characterized in that The training of the network is constrained by the following two-branch joint loss function: in and are the loss functions of the change detection network and the denoising network, respectively. and are the weights of change detection loss and denoising loss respectively.

9. A noise-resistant polarimetric SAR image change detection system, characterized in that: include: Data acquisition module, used to acquire polarimetric SAR images; The change detection module is used to input the acquired polarimetric SAR image into the trained denoising and change detection model, and output the denoising result and the change detection prediction result; wherein the training of the denoising and change detection model includes: The data set is constructed using the covariance matrix components and polarization decomposition components of polarimetric SAR images; Construct a denoising network and a change detection network, wherein the denoising network is a residual convolutional network with embedded channel attention and cross attention, and the change detection network is a transformer network with embedded spatial attention and channel attention; The denoising network is trained using the data set to obtain denoising results of polarimetric SAR images of the same location at different times, and the denoising results are input into the change detection network for training. Meanwhile, the deep features extracted by the denoising network are converted into attention weights and input into the change detection network to recalibrate the recognition capability.

10. A noise-resistant polarimetric SAR image change detection device, characterized in that: The method comprises a memory and a processor, wherein the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the steps of the method according to any one of claims 1 to 8.

Images