Scattering mechanism guided complete polarization SAR data reconstruction method
Through the method of scattering mechanism guidance, the target decomposition network and the fully polarized SAR reconstruction network are used, and the loss function is used to constrain the distance between the reconstructed data and the real data, which solves the low accuracy of the double-polarized SAR reconstruction of the fully polarized SAR data. The problem has achieved higher precision and generalization reconstruction effect.
Patent Information
- Application Number
- CN202411246147.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-12-20
AI Technical Summary
The existing method of reconstructing fully polarized SAR data based on dual-polarized SAR has low accuracy and does not fuse scattering mechanism information, resulting in insufficient generalization and accuracy.
Using the method of scattering mechanism guidance, the scattering component data is obtained through rapid generalized polarized target decomposition, the target decomposition network and the fully polarized SAR reconstruction network are constructed, and the network is trained using the dual-polarized SAR data to design the loss function constraint reconstruction data and reality. The distance of the data and the distance of the scattered components improve the reconstruction accuracy.
It realizes higher precision fully polarized SAR data reconstruction, improves the generalization and accuracy of the reconstruction, and can more effectively reconstruct the fully polarized SAR data.
Smart Images
Figure CN119165485A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of remote sensing imaging, and in particular relates to the technical field of full-polarization SAR imaging remote sensing application. Background Art
[0002] Synthetic Aperture Radar (SAR) is an advanced active microwave sensor that has the advantage of working around the clock and in all weather conditions. SAR has a wide observation coverage and high imaging resolution, and is widely used in earth observation and other aspects.
[0003] The introduction of polarization information has expanded the application scope of SAR images and improved the accuracy of SAR image interpretation. Although full polarization SAR can obtain richer electromagnetic scattering information of the target and can perform mechanical interpretation and structural cognition of the target, due to the influence of factors such as observation width and resolution, the full polarization SAR data obtained by the spaceborne SAR system is relatively small. Compared with full polarization SAR, dual-polarization SAR has the advantage of twice the observation width while maintaining partial polarization information, and is widely used in various SAR systems. Due to the complex implicit mapping relationship between the electromagnetic scattering of targets in different polarization channels, the current methods for reconstructing full polarization SAR data based on dual-polarization SAR are mainly based on reduced polarization. There are few methods for reconstructing full polarization SAR data for general dual-polarization imaging modes. Moreover, these full polarization SAR data reconstruction methods are mainly data-driven and do not integrate scattering mechanism information for reconstruction. Therefore, the generalization and accuracy of these methods are not high. Summary of the invention
[0004] In order to solve the problem that the data reconstruction accuracy of the method for reconstructing full polarization SAR data based on dual-polarization SAR is not high, the present invention provides a scattering mechanism guided full polarization SAR data reconstruction method, the method comprising the following steps: S1. Perform fast generalized polarization target decomposition on the full polarization SAR data to obtain scattered component data and construct a full polarization SAR data-scattering component data set; extract dual-polarization SAR data from the full polarization SAR data and construct a dual-polarization SAR-full polarization SAR paired data set; S2, build the target decomposition network and the full polarization SAR reconstruction network; S3, using the fully polarized SAR data - scattering component data set to train the target decomposition network; S4, using the dual-polarization SAR-full-polarization SAR paired data set to train the full-polarization SAR reconstruction network; S5. Reconstruct the dual-polarization SAR data of other regions and other sensors using the trained full-polarization SAR reconstruction network to obtain reconstructed full-polarization SAR data.
[0005] Furthermore, the scattered component data includes four types: secondary scattered component, odd scattered component, volume scattered component and spiral scattered component.
[0006] Furthermore, the dual-polarization SAR data is divided into three types: HH-VH, VV-HV and HH-VV.
[0007] Furthermore, the target decomposition network includes seven convolutional layers, and a ReLU activation function is set after the first to sixth convolutional layers, the convolution kernel sizes of all convolutional layers are 3×3, the void rates of each convolutional layer are 1, 2, 3, 4, 3, 2 and 1 respectively, the number of input channels of the first convolutional layer is 9 and the number of output channels is 64, the number of input and output channels of the second to sixth convolutional layers are both 64, the number of input channels of the seventh convolutional layer is 64 and the number of output channels is 4; wherein, the output of the first layer and the output of the third layer are used as residuals, and the residual outputs are sent to the fourth convolutional layer, the output of the fourth convolutional layer and the output of the sixth convolutional layer are used as residuals, and the residual outputs are sent to the seventh convolutional layer and then output.
[0008] Furthermore, the fully polarimetric SAR reconstruction network includes seven convolutional layers, and a ReLU activation function is set after the first to sixth convolutional layers, the convolution kernel sizes of all convolutional layers are 3×3, the void rates of each convolutional layer are 1, 2, 3, 4, 3, 2 and 1 respectively, the number of input channels of the first convolutional layer is 4 and the number of output channels is 64, the number of input and output channels of the second to sixth convolutional layers are both 64, the number of input channels of the seventh convolutional layer is 64 and the number of output channels is 5; wherein, the output of the first layer and the output of the third layer are used as residuals, the residual outputs are sent to the fourth convolutional layer, the output of the fourth convolutional layer and the output of the sixth convolutional layer are used as residuals, the residual outputs are sent to the seventh convolutional layer and then output.
[0009] Furthermore, when the target decomposition network is trained using the fully polarized SAR data-scattering component dataset, the loss function is: ; in, ( ) represents the target decomposition network, represents the fully polarized SAR data, represents the scattered component data, Represents the calculation matrix norm, .
[0010] Furthermore, when the dual-polarization SAR-full-polarization SAR paired data set is used to train the full-polarization SAR reconstruction network, the loss function is: ; in, ( ) represents the full polarimetric SAR reconstruction network, represents dual-polarization SAR data, is a hyperparameter and , represents the polarization term that needs to be reconstructed, Indicates that two matrices are concatenated along the feature dimension.
[0011] Further, after extracting dual-polarization SAR data from the full-polarization SAR data, the remaining data is the polarization item that needs to be reconstructed. . Furthermore, the present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned full-polarization SAR data reconstruction method when executing the computer program.
[0012] Furthermore, the present invention also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the above-mentioned full-polarization SAR data reconstruction method.
[0013] The beneficial effects of the method of the present invention are: A scattering mechanism-guided full-polarimetric SAR data reconstruction method is proposed. First, a target decomposition network based on full-polarimetric SAR data is constructed, and the fast generalized polarimetric target decomposition result is used as the true value to train the target decomposition network. Secondly, a full-polarimetric SAR data reconstruction network based on dual-polarimetric SAR is designed. The spatial features rich in context information are extracted from the dual-polarimetric SAR data by using a convolutional neural network fused with residual connections and dilated convolution kernels, and then the spatial features are mapped to the full-polarimetric SAR data. In order to fully train the reconstruction network, in the loss function, not only the distance between the reconstructed full-polarimetric SAR data and the real full-polarimetric SAR data is constrained, but also the scattered components are extracted from the reconstructed and real full-polarimetric SAR data respectively by using the target decomposition network, which further constrains the distance between the reconstructed scattering components and the real scattering components, thereby improving the reconstruction accuracy of the full-polarimetric SAR data. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 is a flow chart of the method of the present invention; Figure 2 Figure (a) is the full-polarization pseudo-color Pauli map of MPOLSAR Hainan region, Figure (b) is the pseudo-color map synthesized by secondary scattering, odd scattering and volume scattering components of MPOLSAR Hainan region, Figure (c) is the VV channel amplitude map of MPOLSAR Hainan region dual-polarization SAR data, and Figure (d) is the HV channel amplitude map of MPOLSAR Hainan region dual-polarization SAR data; Figure 3 Schematic diagram of the target decomposition network structure; Figure 4 In the figure, Figure (a) is the real full-polarization pseudo-color Pauli image of MPOLSAR, Figure (b) is the reconstructed full-polarization pseudo-color Pauli image of the comparison method, and Figure (c) is the reconstructed full-polarization pseudo-color Pauli image of the method described in the present invention. DETAILED DESCRIPTION
[0015] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of the present invention.
[0016] The present invention provides a scattering mechanism guided full polarization SAR data reconstruction method, the flow chart of the method is as follows: Figure 1 As shown, the method of the present invention is described in detail below in conjunction with the flow chart.
[0017] Extract the elements of the fully polarimetric SAR covariance matrix as a vector:
[0018] in is the fully polarized SAR data, It represents the target scattering coefficient when the radar transmits electromagnetic waves in the vertical polarization direction and receives electromagnetic waves in the horizontal polarization direction. It represents the target scattering coefficient when the radar transmits electromagnetic waves in the horizontal polarization direction and receives electromagnetic waves in the horizontal polarization direction. It represents the target scattering coefficient when the radar receives electromagnetic waves in the vertical and horizontal polarization directions. It means taking conjugate of variables. It means to calculate the absolute value of an absolute complex number. represents the operation of calculating logarithms, Indicates the calculation of the phase of a complex number. Figure 2 (a) shows the full polarimetric pseudo-color Pauli image of the MPOLSAR Hainan region.
[0019] By performing fast generalized polarimetric target decomposition on the fully polarimetric SAR data, the secondary scattering, odd scattering, volume scattering and spiral scattering components are obtained as follows: ; in is the set of scattered components, represents the secondary scattering component, represents the odd-order scattered component, represents the volume scattering component, represents the spiral scattering component. Figure 2 (b) shows a pseudo-color image of the MPOLSAR Hainan region synthesized from secondary scattering, odd scattering and volume scattering components.
[0020] The fast generalized polarimetric target decomposition is a technology for processing SAR data. For specific technical content, please refer to the paper "Polarimetric Model-Based Decomposition with Refined Double-BounceOrientation Angle and Scattering Model" (Remote Sens. 2021, 13, 3070.).
[0021] There are three dual polarization modes: HH-VH, VV-HV and HH-VV. Taking the VV-HV dual-polarization mode as an example, the dual-polarization SAR data extracted from the full-polarization SAR data is: ; in The VV channel amplitude diagram and HV channel amplitude diagram of the VV-VH dual-polarization SAR data extracted from it are shown as follows: Figure 2 (c) and Figure 2 (d) as shown.
[0022] After extracting dual-polarization SAR data from the full set of SAR data, the remaining data is the polarization item that needs to be reconstructed. Therefore, the dual-polarization data extracted each time is different, and the corresponding polarization items that need to be reconstructed are also different. Taking the VV-HV dual-polarization mode as an example, the dual-polarization SAR data Compared with full polarimetric SAR data , the polarization term that needs to be reconstructed is:
[0023] The above method is used to construct the full polarization SAR data-scattering component data set and the dual-polarization SAR-full polarization SAR paired data set.
[0024] In order to facilitate the subsequent training of the neural network, , , and The slice images are cropped into 40×40 size and then packaged into a dataset to train the target decomposition network and the full polarization SAR reconstruction network.
[0025] By using a neural network with fused residual connections and dilated convolutions as the target decomposition network , feature extraction is performed on dual-polarization SAR data.
[0026] like Figure 3 As shown, It consists of seven convolutional layers, and ReLU activation functions are set after the first to sixth convolutional layers. The convolution kernel size of all convolutional layers is 3×3, and the number of input channels and output channels of the intermediate layers (second to sixth layers) are both 64. The first convolutional layer has 9 input channels and 64 output channels, and uses a convolution kernel with a dilation rate of 1. After the input full-polarization SAR data passes through the first convolution and ReLU activation function, it passes through the second and third convolutional layers in turn, using convolution kernels with dilation rates of 2 and 3 respectively, where the output of the first layer and the output of the third layer are used as residuals. The residual output is sent to the fourth convolutional layer, which uses a convolution kernel with a dilation rate of 4, and then the output of the fourth convolutional layer passes through the fifth and sixth convolutional layers with dilation rates of 3 and 2, and then the output of the fourth convolutional layer and the output of the sixth convolutional layer are used as residuals. The residual output is output through the last convolution layer, where the last convolution layer only includes convolution operations, its hole rate is 1, the number of input channels is 64, and the number of output channels is 4.
[0027] Fully polarimetric SAR data Input target decomposition network , the output of the network With the scattered component collection The loss function is: ; in, Represents the calculation matrix norm, .
[0028] The network optimizer selects the Adaptive Moment Estimation (ADAM) optimizer. The optimizer parameters are set to , as well as , the learning rate of the network is set to 0.001.
[0029] Full Polarimetric SAR Reconstruction Network Decomposition network with target Except that the number of input channels in the first layer is 4 and the number of output channels in the last layer is 5, the rest of the network structures are the same.
[0030] The dual-polarization SAR data Input full polarimetric SAR reconstruction network , the output of the network and the polarization term that needs to be reconstructed The loss function is: ; in is a hyperparameter, ; Indicates that the two matrices are concatenated in the feature dimension. The first term of the loss function minimizes the distance between the output of the full polarization SAR reconstruction network and the polarization term to be reconstructed, and the second term minimizes the distance between the reconstructed scattering component and the real scattering component.
[0031] The network optimizer selects the Adaptive Moment Estimation (ADAM) optimizer. The optimizer parameters are set to , as well as , the learning rate of the network is set to 0.001.
[0032] Finally, the dual-polarization SAR data from other regions and other sensors are input into the trained full-polarization SAR reconstruction network. The reconstruction results of fully polarized SAR data are obtained.
[0033] The method of the present invention not only constrains the distance between the reconstructed polarization term and the real polarization term, but also constrains the distance between the network reconstructed scattering component and the real scattering component by designing a loss function guided by the fusion scattering mechanism, which can more effectively reconstruct the full polarization SAR data and has high practicality.
[0034] In order to verify the effectiveness and practical applicability of the method described in the present invention, the comparison method is set to use only the loss function The first one in trains the fully polarimetric SAR reconstruction network.
[0035] Figure 4 (a) shows the real MPOLSAR full polarization pseudo-color Pauli image. Figure 4 (b) and Figure 4 (c) shows the reconstructed full-polarization pseudo-color Pauli images of the comparison method and the method of the present invention, respectively. It can be seen that the reconstruction effect of the comparison method in areas such as forests, bridges and buildings is not as good as that of the method of the present invention. The full-polarization pseudo-color Pauli image reconstructed by the method of the present invention is more similar to the real full-polarization pseudo-color Pauli image.
[0036] The root mean square error (RMS) index is used to evaluate the accuracy of different reconstruction methods. Table 1 shows the RMS errors of different reconstruction items for different reconstruction methods. It can be seen that the method described in the present invention has , , and The reconstruction accuracy of the term is higher than that of the comparison method. The reconstruction accuracy of the item is slightly lower.
[0037] In summary, the method of this patent can more accurately reconstruct dual-polarization SAR data into full-polarization SAR data.
[0038] Table 1:
Claims
1. A scattering mechanism guided full polarization SAR data reconstruction method, characterized in that: The method comprises the following steps: S1. Perform fast generalized polarization target decomposition on the full polarization SAR data to obtain scattered component data and construct a full polarization SAR data-scattering component data set; extract dual-polarization SAR data from the full polarization SAR data and construct a dual-polarization SAR-full polarization SAR paired data set; S2, build the target decomposition network and the full polarization SAR reconstruction network; S3, using the fully polarized SAR data - scattering component data set to train the target decomposition network; S4, using the dual-polarization SAR-full-polarization SAR paired data set to train the full-polarization SAR reconstruction network; S5. Reconstruct the dual-polarization SAR data of other regions and other sensors using the trained full-polarization SAR reconstruction network to obtain reconstructed full-polarization SAR data.
2. The method for reconstructing fully polarized SAR data according to claim 1, characterized in that: The scattered component data includes four types: secondary scattered component, odd scattered component, volume scattered component and spiral scattered component.
3. The method for reconstructing fully polarized SAR data according to claim 2, characterized in that: The dual-polarization SAR data are divided into three types: HH-VH, VV-HV and HH-VV.
4. The method for reconstructing fully polarized SAR data according to claim 3, characterized in that: The target decomposition network includes seven convolutional layers, and ReLU activation functions are set after the first to sixth convolutional layers, the convolution kernel sizes of all convolutional layers are 3×3, the void rates of each convolutional layer are 1, 2, 3, 4, 3, 2 and 1 respectively, the number of input channels of the first convolutional layer is 9 and the number of output channels is 64, the number of input and output channels of the second to sixth convolutional layers are both 64, the number of input channels of the seventh convolutional layer is 64 and the number of output channels is 4; wherein, the output of the first layer and the output of the third layer are used as residuals, the residual outputs are sent to the fourth convolutional layer, the output of the fourth convolutional layer and the output of the sixth convolutional layer are used as residuals, the residual outputs are sent to the seventh convolutional layer and then output.
5. The method for reconstructing fully polarized SAR data according to claim 4, characterized in that: The fully polarimetric SAR reconstruction network includes seven convolutional layers, and ReLU activation functions are arranged after the first to sixth convolutional layers, the convolution kernel sizes of all convolutional layers are 3×3, the hole rates of each convolutional layer are 1, 2, 3, 4, 3, 2 and 1 respectively, the number of input channels of the first convolutional layer is 4 and the number of output channels is 64, the number of input and output channels of the second to sixth convolutional layers are both 64, the number of input channels of the seventh convolutional layer is 64 and the number of output channels is 5; wherein, the output of the first layer and the output of the third layer are used as residuals, the residual outputs are sent to the fourth convolutional layer, the output of the fourth convolutional layer and the output of the sixth convolutional layer are used as residuals, the residual outputs are sent to the seventh convolutional layer and then output.
6. The method for reconstructing fully polarized SAR data according to claim 5, characterized in that: When using the fully polarized SAR data-scattering component dataset to train the target decomposition network, the loss function is: ； in, ( ) represents the target decomposition network, represents the fully polarized SAR data, represents the scattered component data, Represents the calculation matrix norm, .
7. The method for reconstructing fully polarized SAR data according to claim 6, characterized in that: When the dual-polarization SAR-full-polarization SAR paired data set is used to train the full-polarization SAR reconstruction network, the loss function is: ； in, ( ) represents the full polarimetric SAR reconstruction network, represents dual-polarization SAR data, is a hyperparameter and , represents the polarization term that needs to be reconstructed, Indicates that two matrices are concatenated along the feature dimension.
8. The method for reconstructing fully polarized SAR data according to claim 7, characterized in that: After extracting dual-polarization SAR data from the full-polarization SAR data, the remaining data is the polarization item that needs to be reconstructed. .
9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.
10. A computer-readable storage medium for storing computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.