Isar coherent translational motion compensation and imaging method based on motion parameter estimation network

Abstract

The application discloses an ISAR phase correlation translational motion compensation and imaging method based on a motion parameter estimation network, relates to the field of imaging technology, and comprises the following steps: receiving echo of a spatial target inverse synthetic aperture radar; pre-processing the echo to obtain a compensated echo sequence; processing the compensated echo sequence by using a trained motion parameter estimation network model to obtain a distance grid offset coarse estimation value of a target center position of each echo; multiplying the distance grid offset coarse estimation value by a distance resolution unit to obtain a distance coarse estimation value, fitting the distance coarse estimation value to obtain a translational motion parameter coarse estimation value, taking the translational motion parameter coarse estimation value as an optimized imaging parameter of a preset imaging model, iteratively optimizing the translational motion parameter coarse estimation value to obtain an optimized imaging parameter; obtaining translational motion error terms of each echo according to the optimized imaging parameter; compensating the translational motion error terms of each echo to obtain an imaging result by coherently accumulating the compensated echoes.

Description

Technical Field

[0001] This invention belongs to the field of imaging technology, specifically relating to an ISAR coherent translational compensation and imaging method based on a motion parameter estimation network. Background Technology

[0002] Inverse Synthetic Aperture Radar (ISAR) possesses all-weather, all-day, long-range, and high-resolution imaging capabilities, making it valuable for applications in situational awareness, early warning and control, and civil aviation control. During ISAR imaging, the motion of non-cooperative targets can be equivalently decomposed into two processes: translation along the line-of-sight direction and rotation around the turntable center. The translation along the line-of-sight direction is not only meaningless for ISAR imaging but also introduces envelope shift and phase error, resulting in poor image quality. The process of compensating for this is called translational compensation.

[0003] Traditional translational compensation methods generally consist of two steps: envelope alignment and autofocus. Average Range Profile Energy (ARPE) and Entire Image Contrast (EIC) are typically chosen as the fitness functions for envelope alignment and phase autofocus, respectively. However, envelope offset estimation using ARPE as the objective function often fails to meet the range resolution requirements of focused imaging, and the envelope offset estimation error cannot be compensated for in subsequent phase error estimation, thus limiting the overall algorithm's estimation accuracy. At low signal-to-noise ratios, the performance of most envelope alignment algorithms degrades, leading to the risk of failure for some autofocus algorithms. Furthermore, with the increase in signal bandwidth, the size of the observed target, and the further increase in the imaging accumulation angle, the phase compensation optimization problem exhibits high nonlinearity and complexity. Bionic intelligent optimization algorithms suffer from drawbacks such as high computational dimensionality, computational complexity, excessive parameters, and low estimation accuracy, and these algorithms risk getting trapped in local optima.

[0004] Therefore, there is an urgent need to provide an ISAR coherent translational compensation and imaging method to improve the shortcomings of existing technologies. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention provides an ISAR coherent translational compensation and imaging method based on a motion parameter estimation network. The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] This invention provides an ISAR coherent translational compensation and imaging method based on a motion parameter estimation network, comprising:

[0007] Receive echoes from inverse synthetic aperture radar of space targets;

[0008] The echoes are preprocessed to obtain the compensated echo sequence;

[0009] The trained motion parameter estimation network model is used to process the compensated echo sequence to obtain a coarse estimate of the distance grid offset of the target center position in each echo.

[0010] The coarse estimate of the distance grid offset is multiplied by the distance resolution cell to obtain the coarse distance estimate; the coarse distance estimate is then fitted to obtain the coarse estimate of the translation parameters.

[0011] The coarse estimate of the translational parameters is used as the imaging parameters to be optimized in the preset imaging model. Iterative optimization is performed to obtain the optimized imaging parameters. Based on the optimized imaging parameters, the translational error term of each echo is obtained. The translational error term of each echo is compensated, and the imaging result is obtained by coherently accumulating the compensated echoes.

[0012] The beneficial effects of this invention are:

[0013] This invention provides an ISAR coherent translational compensation and imaging method based on a motion parameter estimation network. The method uses a trained motion parameter estimation network model to process the compensated echo sequence, obtaining a coarse estimate of the range grid offset of the target center position for each echo. This avoids the problem of unadaptive initial value settings in parameterization methods and improves the convergence speed and robustness of subsequent optimization algorithms. The coarse estimate of the range grid offset is multiplied by the range resolution cell to obtain a coarse range estimate. This coarse range estimate is then fitted to obtain a coarse estimate of the translational parameters. This coarse estimate of the translational parameters is used as the imaging parameters to be optimized in a preset imaging model, and iterative optimization is performed to obtain optimized imaging parameters. Based on the optimized imaging parameters, the translational error term for each echo is obtained. The translational error term for each echo is compensated, and the compensated echoes are coherently accumulated to obtain the imaging result. In this invention, the motion parameter estimation process targets the image entropy of the final imaging result. The translational compensation term is directly constructed to jointly compensate for the envelope error and phase error caused by the translation of the target. This avoids the propagation of different parameter estimation errors in the cascaded compensation method, which leads to the accumulation of errors in the final compensation result, and improves the focusing accuracy and image calibration accuracy of the imaging result.

[0014] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0015] Figure 1 This is a flowchart of an ISAR coherent translational compensation and imaging method based on a motion parameter estimation network provided in an embodiment of the present invention;

[0016] Figure 2 This is another flowchart of the ISAR coherent translational compensation and imaging method based on motion parameter estimation network provided in the embodiments of the present invention;

[0017] Figure 3 This is a schematic diagram of a trained motion parameter estimation network model provided in an embodiment of the present invention;

[0018] Figure 4 This is a schematic diagram of a single-layer motion parameter estimation unit provided in an embodiment of the present invention;

[0019] Figure 5 This is a schematic diagram of a multi-layer motion parameter estimation unit provided in an embodiment of the present invention;

[0020] Figure 6 This is a schematic diagram of Mamba provided in an embodiment of the present invention;

[0021] Figure 7 This is a schematic diagram illustrating the comparison of imaging results under different networks provided in an embodiment of the present invention;

[0022] Figure 8 This is a schematic diagram comparing the imaging results of the method of the present invention under different signal-to-noise ratios provided in the embodiments of the present invention. Detailed Implementation

[0023] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0024] Please see Figure 1 , Figure 1 This is a flowchart of an ISAR coherent translational compensation and imaging method based on a motion parameter estimation network provided in an embodiment of the present invention. Figure 2 This is another flowchart of the ISAR coherent translational compensation and imaging method based on a motion parameter estimation network provided in this invention. The ISAR coherent translational compensation and imaging method based on a motion parameter estimation network provided by this invention includes:

[0025] S101, Receive the echo from the inverse synthetic aperture radar of space targets.

[0026] Specifically, in this embodiment, the echo of the space target inverse synthetic aperture radar can be the echo of a large bandwidth, long time inverse synthetic aperture radar.

[0027] S102. Preprocess the echo to obtain the compensated echo sequence.

[0028] Specifically, this embodiment includes:

[0029] The echo is divided into several sub-apertures;

[0030] The delinear frequency modulation method is used to perform pulse compression processing on several sub-apertures to obtain a one-dimensional range image, and the range dimension of the one-dimensional range image is transformed to the frequency domain to obtain the frequency domain signal;

[0031] The compensation factors for the Residual Video Phase (RVP) term and the envelope skew term are constructed, and phase compensation is performed at the peak frequency of each target scattering point in the frequency domain signal to obtain the compensated echo sequence.

[0032] S103. The trained motion parameter estimation network model is used to process the compensated echo sequence to obtain a rough estimate of the distance grid offset of the target center position of each echo.

[0033] Specifically, please see Figure 3 , Figure 3 This is a schematic diagram of a trained motion parameter estimation network model provided in an embodiment of the present invention. In this embodiment, the trained motion parameter estimation network model includes a forward motion parameter estimation module, a backward motion parameter estimation module, a hidden layer, and a stitching layer. The trained motion parameter estimation network model is used to process the compensated echo sequence to obtain a coarse estimate of the distance grid offset of the target center position of each echo, including:

[0034] The compensated echo sequence is converted into multiple vectors arranged in sequence;

[0035] Regarding the first The i-th vector, using the i-th vector The forward motion parameter estimation module estimates the first... The vector and the first The features output by the forward motion parameter estimation module are processed to obtain the first... The features output by the first forward motion parameter estimation module; using the first The backward motion parameter estimation module estimates the first... The vector and the first The features output by the backward motion parameter estimation module are processed to obtain the first... The features output by the backward motion parameter estimation module; using the features of the first backward motion parameter estimation module; The hidden layer for the first The features output by the first forward motion parameter estimation module and the first The features output by the backward motion parameter estimation module are processed to obtain the first... Features output by each hidden layer;

[0036] A stitching layer is used to stitch together the features output by all hidden layers to obtain a coarse estimate of the distance grid offset of the target center position for each echo.

[0037] Please continue reading Figure 3 , , , and Represents multiple vectors arranged sequentially, xM-mb b (xLSTM-Mamba b () represents the forward motion parameter estimation module, xM-mb f (xLSTM-Mamba f () indicates the backward motion parameter estimation module. , , and This indicates a hidden layer, while Full Connection indicates a spliced ​​layer.

[0038] In this embodiment, the forward motion parameter estimation module includes a cascaded first single-layer motion parameter estimation unit, a second single-layer motion parameter estimation unit, a first multi-layer motion parameter estimation unit, a third single-layer motion parameter estimation unit, and a fourth single-layer motion parameter estimation unit. The first, second, third, and fourth single-layer motion parameter estimation units have the same structure, each including a first normalization layer, a first causal convolutional layer, a first block diagonal structured long short-term memory network, a first set of normalization layers, a first projection layer, a first Gaussian error linearization unit, a first residual connection layer, and a first Mamba layer, all connected in sequence. The first multi-layer motion parameter estimation unit includes a second normalization layer, a second projection layer, a second causal convolutional layer, a first neural network combining a block diagonal matrix and a long short-term memory network, a second set of normalization layers, a first neural network using the Swish activation function, a second residual connection layer, and a second Mamba layer, all connected in sequence. The forward motion parameter estimation module estimates the first... The vector and the first The features output by the forward motion parameter estimation module are processed to obtain the first... The features output by each forward motion parameter estimation module include:

[0039] The first single-layer motion parameter estimation unit, the second single-layer motion parameter estimation unit, the first multi-layer motion parameter estimation unit, the third single-layer motion parameter estimation unit, and the fourth single-layer motion parameter estimation unit are cascaded together to estimate the motion parameter parameters of the first single-layer motion parameter estimation unit. The vector and the first The features output by the forward motion parameter estimation module are processed to obtain the first... Features output by the forward motion parameter estimation module.

[0040] In this embodiment, the backward motion parameter estimation module includes a cascaded fifth single-layer motion parameter estimation unit, a sixth single-layer motion parameter estimation unit, a second multi-layer motion parameter estimation unit, a seventh single-layer motion parameter estimation unit, and an eighth single-layer motion parameter estimation unit. The fifth, sixth, seventh, and eighth single-layer motion parameter estimation units have the same structure, each including a third normalization layer, a third causal convolutional layer, a second diagonal structured long short-term memory network, a third set of normalization layers, a third projection layer, a second Gaussian error linearization unit, a third residual connection layer, and a third Mamba layer, all connected in sequence. The second multi-layer motion parameter estimation unit includes a fourth normalization layer, a fourth projection layer, a fourth causal convolutional layer, a second neural network combining a diagonal matrix and a long short-term memory network, a fourth set of normalization layers, a second neural network using the Swish activation function, a fourth residual connection layer, and a fourth Mamba layer, all connected in sequence. The backward motion parameter estimation module estimates the first... The vector and the first The features output by the backward motion parameter estimation module are processed to obtain the first... The features output by the backward motion parameter estimation module include:

[0041] The fifth, sixth, second, seventh, and eighth single-layer motion parameter estimation units are cascaded together to estimate the motion parameters of the third single-layer motion parameter estimation unit. The vector and the first The features output by the backward motion parameter estimation module are processed to obtain the first... Features output by the backward motion parameter estimation module.

[0042] It should be noted that the first single-layer motion parameter estimation unit, the second single-layer motion parameter estimation unit, the third single-layer motion parameter estimation unit, the fourth single-layer motion parameter estimation unit, the fifth single-layer motion parameter estimation unit, the sixth single-layer motion parameter estimation unit, the seventh single-layer motion parameter estimation unit, and the eighth single-layer motion parameter estimation unit in this embodiment have the same structure, and the first multi-layer motion parameter estimation unit and the second multi-layer motion parameter estimation unit have the same structure.

[0043] Please see Figure 4 and Figure 5 and combined Figure 3 , Figure 4 This is a schematic diagram of a single-layer motion parameter estimation unit provided in an embodiment of the present invention. Figure 5This is a schematic diagram of a multi-layer motion parameter estimation unit provided in an embodiment of the present invention. The motion parameter estimation network model provided by the present invention is composed of a bidirectional extended long short-term memory network (xLSTM) combined with the Mamba architecture. It consists of a forward motion parameter estimation module (forward xLSTM-Mamba) and a backward motion parameter estimation module (backward xLSTM-Mamba) to achieve bidirectional processing. Both the forward xLSTM-Mamba and the backward xLSTM-Mamba include multiple single-layer motion parameter estimation units (single-layer LSTM-Mamba) and one multi-layer motion parameter estimation unit (multi-layer multi-LSTM-Mamba). The single-layer motion parameter estimation unit includes a normalization layer 1, a causal convolution layer 2, a block diagonal structured long short-term memory network 3 (Block Diagonal sLSTM), a group normalization layer 4 (Group Normalization), projection layers 5, Gaussian error linearization units 6 (GELU Layer), and a residual connection layer 7 (Residual). The multilayer motion parameter estimation unit comprises, in sequence, a layer normalization layer (9), projection layers (10), causal convolution layers (11), a neural network (12) combining a block diagonal matrix and a long short-term memory network (12), a group normalization layer (13), a neural network (14) using the Swish activation function (14), a residual connection layer (15), and Mamba (16).

[0044] It should be noted that the Mamba architecture uses the existing State Space Model (SSM) framework; the Swish activation function is the existing Swish.

[0045] In this embodiment, please refer to Figure 6 , Figure 6 This is a schematic diagram of Mamba provided in an embodiment of the present invention. The first Mamba, second Mamba, third Mamba, and fourth Mamba have the same structure, each including: a first linear layer 17, a second linear layer 18, a convolutional layer 19, a first swish activation function 20, a second swish activation function 21, a state-space model 22, a third linear layer 23, and a fully connected layer 24; the processing of the first Mamba, second Mamba, third Mamba, or fourth Mamba includes:

[0046] The first linear layer 17 is used to process the input features to obtain the first feature; the first swish activation function 20 is used to process the first feature to obtain the second feature;

[0047] The input features are processed by the second linear layer 18 to obtain the third feature; the third feature is processed by the convolutional layer 19 to obtain the fourth feature; and the fourth feature is processed by the second swish activation function 21 to obtain the fifth feature.

[0048] The fifth feature is processed using state-space model 22 to obtain the sixth feature; the sixth feature is multiplied by the second feature to obtain the seventh feature;

[0049] The seventh feature is processed using the third linear layer 23 to obtain the eighth feature;

[0050] The eighth feature is processed using a fully connected layer 24 to obtain the output feature.

[0051] In this embodiment, the training process of the trained motion parameter estimation network model includes:

[0052] In the In the next iteration, data similar to the compensated echo sequence is constructed and used as training data. The initial state of the motion parameter estimation network model to be trained The true distance parameter of the compensated echo sequence is used as the state. It should be noted that the training data is constructed based on the compensated echo sequence to form the training dataset.

[0053] Get the The action corresponding to the next motion parameter estimation network model to be trained is expressed as follows: , This represents the action sampled from the distribution;

[0054] Calculate the initial state With state The gap loss is expressed as: , Indicates the gap loss, according to the first The gap loss, actions, and learning rate from the training process are backpropagated to update the training data for the second training cycle. The parameters of the motion parameter estimation network model to be trained are expressed as follows: , These represent the parameters of the motion parameter estimation network model. This represents the learning rate used to train the motion parameter estimation network model. Represents the loss function gradient, This represents the trained motion parameter estimation network model, yielding the first... The next training iteration of the motion parameter estimation network model;

[0055] This process is iterated until the number of training iterations or the degree of convergence meets the preset iteration conditions, resulting in a well-trained motion parameter estimation network model. Table 1 describes the detailed training process. Indicates the number of training iterations. This indicates the total number of times the training dataset has been trained.

[0056] Table 1 Training Strategies

[0057]

[0058] S104. Multiply the coarse estimate of the distance grid offset by the resolution cell to obtain the coarse estimate of the distance; fit the coarse estimate of the distance to obtain the coarse estimate of the translation parameters.

[0059] This invention utilizes the relationship between the echo after target pulse compression and translational parameters, and employs a trained motion parameter estimation network model to coarsely estimate the translational parameters. This avoids the problem of the initial value setting being unable to adapt in parameterization methods, and at the same time improves the convergence speed and robustness of subsequent optimization algorithms.

[0060] Specifically, in this embodiment, the coarse estimate of the translational parameters includes velocity. acceleration and accelerometer .

[0061] It should be noted that, according to the basic principles of ISAR imaging, the range resolution unit is defined as... , Represents the speed of light. Indicates bandwidth.

[0062] S105. The coarse estimate of the translational parameters is used as the imaging parameters to be optimized in the preset imaging model. Iterative optimization is performed to obtain the optimized imaging parameters. Based on the optimized imaging parameters, the translational error term of each echo is obtained. The translational error term of each echo is compensated, and the imaging result is obtained by coherently accumulating the compensated echoes.

[0063] Specifically, in this embodiment, it further includes: according to preset conditions, making the preset imaging model reach the iteration termination condition, obtaining the optimal imaging parameters, and obtaining the corresponding imaging result.

[0064] In this embodiment, the preset condition includes that the difference between the loss function value of the current iteration and the loss function value of the previous iteration is less than a preset value, such as... Figure 2 As shown, determine If the condition is not met, continue iterative updates. Optional, default value The expression for the loss function value in the current iteration is:

[0065] ;

[0066] ;

[0067] in, Represents the image entropy function. Indicates the azimuth length of the image. Indicates the distance length of the image. Indicates the orientation index of the image. Indicates the image distance index. This represents the mean of the imaging results in the current iteration. This indicates the imaging result of the current iteration.

[0068] In this invention, the imaging parameter estimation process targets the image entropy of the imaging result. It directly constructs translational compensation to compensate for the envelope error and phase error caused by the translation of the target, avoiding the accumulation of errors in the compensation result caused by the propagation of different parameter estimation errors in cascaded compensation, and improving the focus and image calibration accuracy of the imaging result.

[0069] In this embodiment, the coarse estimate of the translational parameters is used as the imaging parameters to be optimized in the preset imaging model, and iterative optimization is performed to obtain the optimized imaging parameters, including:

[0070] The compensated echo sequence is used as the imaging parameters to be optimized. Initialize first-order moments and second moment =0;

[0071] Calculate the gradient of the loss function value in the current iteration. Its expression is:

[0072] ;

[0073] in, Represents the gradient. Represents the image entropy function. This represents the imaging parameters to be optimized in the current iteration. This represents the imaging result of the current iteration. Indicates the orientation index of the image. Indicates the image distance index;

[0074] The updated estimate of the first moment is expressed as follows:

[0075] ;

[0076] The estimated value of the second moment is updated, and its expression is:

[0077] ;

[0078] The updated estimates of the first and second moments are corrected for bias to obtain the corrected estimates of the first moment. and the estimated value of the corrected second moment Their expressions are as follows:

[0079] ;

[0080] ;

[0081] Based on the corrected estimate of the first moment and the estimated value of the corrected second moment The optimized imaging parameter estimates after the current iteration are obtained, and their expression is:

[0082] ;

[0083] in, Indicates the learning rate. and These represent the decay rates of the first-order and second-order moment estimates, respectively (typically close to 1, for example...). , ), To represent a constant (usually a very small number to avoid division by zero, it can be set to...), ).

[0084] Optionally, this invention employs the Adam optimization algorithm to accurately estimate the target translational parameters. The Adam optimization algorithm is suitable for non-convex optimization problems, escapes saddle points quickly, and converges faster than other gradient descent algorithms. Furthermore, the parameters are unaffected by gradient scaling and can automatically adjust the learning rate, making it suitable for unstable objective functions. In addition, using image entropy as the optimization objective function allows for the simultaneous acquisition of energy gains from the coherent accumulation of both distance and orientation dimensions, ensuring the optimization algorithm has high noise robustness.

[0085] In summary, this invention provides an ISAR coherent translational compensation and imaging method based on a motion parameter estimation network. Utilizing the relationship between the target pulse compression echo and translational parameters, a motion parameter estimation network model is used to coarsely estimate the translational parameters, avoiding the problem of unadaptive initial value settings in parameterization methods. This also improves the convergence speed and robustness of subsequent optimization algorithms. The motion parameter estimation process targets the image entropy of the final imaging result, directly constructing a translational compensation term to jointly compensate for the envelope and phase errors caused by target translation. This avoids the propagation of errors from different parameter estimation methods in cascaded compensation methods, leading to error accumulation in the final compensation result, thus improving the focus and image calibration accuracy of the imaging result. Furthermore, the Adam optimization algorithm, used for non-convex optimization problems, escapes saddle points quickly, has a faster convergence speed than other gradient descent algorithms, and its parameter updates are unaffected by gradient scaling transformations, automatically adjusting the learning rate, making it suitable for unstable objective functions. Simultaneously, using image entropy as the optimization objective function allows for the simultaneous acquisition of energy gains from the coherent accumulation of range and azimuth dimensions, ensuring high noise robustness of the algorithm.

[0086] In an optional embodiment of the present invention, the effectiveness of the ISAR coherent translational compensation and imaging method based on motion parameter estimation network provided in the above embodiment is verified by simulation experiments, specifically as follows:

[0087] Please see Figure 7 , Figure 7 This is a schematic diagram comparing imaging results under different networks provided in this invention. Various methods exhibit different performance characteristics in estimating and compensating motion parameters. These methods include deep learning models with self-attention mechanisms (Transformer), Long Short-Term Memory (LSTM) networks, Bi-xLSTM networks, the Mamba architecture, and the proposed Bi-xLSTM combined with Mamba architecture (Bi-xLSTM-Mamba). The magnified view shows that the estimated curve obtained by the proposed Bi-xLSTM-Mamba network has the smallest error compared to the true value, and the final image is more focused. This further illustrates that the network proposed in this invention not only provides more accurate initial estimates but also effectively refines these estimates through an optimization process, thereby obtaining superior motion-compensated imaging results.

[0088] It should be noted that, Figure 7 The label represents the theoretical motion parameter curve, esti represents the estimated motion parameter curve, the first row represents the initial estimation result, the second row represents the optimized estimation result, and the last row represents the imaging result.

[0089] Please see Figure 8 , Figure 8 This diagram illustrates a comparison of imaging results of the method proposed in this invention under different signal-to-noise ratios (SNRs), namely 0dB, 5dB, 10dB, and 15dB. The proposed method exhibits good robustness at various SNR levels. The initial estimation results in the first row show that even under low SNR conditions, the estimation results have a certain degree of accuracy, although there is a slight deviation from the true values. After Adam optimization (second row), the estimation results are significantly improved, especially with a significant reduction in error under low SNR conditions. The final imaging results (third row) maintain clear details even with gradually increasing noise, demonstrating the effectiveness of the proposed method in maintaining high-quality imaging under different SNR conditions.

[0090] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not expressly listed. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device comprising said element. Terms such as "connected" or "linked" are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect. The orientations or positional relationships indicated by terms such as "upper," "lower," "left," and "right" are based on the orientations or positional relationships shown in the accompanying drawings and are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention.

[0091] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0092] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A coherent translational compensation and imaging method for ISAR based on a motion parameter estimation network, characterized in that, include: Receive echoes from inverse synthetic aperture radar of space targets; The echoes are preprocessed to obtain a compensated echo sequence; The compensated echo sequence is processed using a trained motion parameter estimation network model to obtain a coarse estimate of the distance grid offset of the target center position for each echo. The coarse estimate of the distance grid offset is multiplied by the distance resolution cell to obtain a coarse distance estimate; the coarse distance estimate is then fitted to obtain a coarse estimate of the translation parameters. The coarse estimate of the translational parameters is used as the imaging parameters to be optimized in the preset imaging model. Iterative optimization is performed to obtain the optimized imaging parameters. Based on the optimized imaging parameters, the translational error terms of each echo are obtained. The translational error terms of each echo are compensated, and the imaging results are obtained by coherently accumulating the compensated echoes. The trained motion parameter estimation network model includes a forward motion parameter estimation module, a backward motion parameter estimation module, a hidden layer, and a stitching layer. The process of using the trained motion parameter estimation network model to process the compensated echo sequence to obtain a coarse estimate of the distance grid offset of the target center position for each echo includes: The compensated echo sequence is converted into multiple vectors arranged in sequence; Regarding the first The i-th vector, using the i-th vector The forward motion parameter estimation module estimates the first... The vector and the first The features output by the forward motion parameter estimation module are processed to obtain the first... The features output by the first forward motion parameter estimation module; using the first The backward motion parameter estimation module estimates the first... The vector and the first The features output by the backward motion parameter estimation module are processed to obtain the first... The features output by the backward motion parameter estimation module; using the features of the first backward motion parameter estimation module; The first hidden layer is for the first... The features output by the first forward motion parameter estimation module and the first The features output by the backward motion parameter estimation module are processed to obtain the first... Features output by each hidden layer; A stitching layer is used to stitch together the features output by all hidden layers to obtain a coarse estimate of the distance grid offset of the target center position for each echo.

2. The ISAR coherent translational compensation and imaging method based on motion parameter estimation network according to claim 1, characterized in that, The preprocessing of the echo to obtain the compensated echo sequence includes: The echo is divided into several sub-apertures; The delinear frequency modulation method is used to perform pulse compression processing on the several sub-apertures to obtain a one-dimensional range image, and the range dimension of the one-dimensional range image is transformed to the frequency domain to obtain a frequency domain signal; Compensation factors for residual video terms and envelope skew terms are constructed, and phase compensation is performed at the peak frequency of each target scattering point in the frequency domain signal to obtain the compensated echo sequence.

3. The ISAR coherent translational compensation and imaging method based on motion parameter estimation network according to claim 1, characterized in that, The forward motion parameter estimation module includes a cascaded first single-layer motion parameter estimation unit, a second single-layer motion parameter estimation unit, a first multi-layer motion parameter estimation unit, a third single-layer motion parameter estimation unit, and a fourth single-layer motion parameter estimation unit. The first, second, third, and fourth single-layer motion parameter estimation units have the same structure, each including a first normalization layer, a first causal convolutional layer, a first block diagonal structured long short-term memory network, a first set of normalization layers, a first projection layer, a first Gaussian error linearization unit, a first residual connection layer, and a first Mamba layer, all connected in sequence. The first multi-layer motion parameter estimation unit includes a second normalization layer, a second projection layer, a second causal convolutional layer, a first neural network combining a block diagonal matrix and a long short-term memory network, a second set of normalization layers, a first neural network using the Swish activation function, a second residual connection layer, and a second Mamba layer, all connected in sequence. The forward motion parameter estimation module estimates the first... The vector and the first The features output by the forward motion parameter estimation module are processed to obtain the first... The features output by each forward motion parameter estimation module include: The first single-layer motion parameter estimation unit, the second single-layer motion parameter estimation unit, the first multi-layer motion parameter estimation unit, the third single-layer motion parameter estimation unit, and the fourth single-layer motion parameter estimation unit are cascaded together to estimate the motion parameter parameters of the first single-layer motion parameter estimation unit. The vector and the first The features output by the forward motion parameter estimation module are processed to obtain the first... Features output by the forward motion parameter estimation module.

4. The ISAR coherent translational compensation and imaging method based on motion parameter estimation network according to claim 3, characterized in that, The backward motion parameter estimation module includes a cascaded fifth single-layer motion parameter estimation unit, a sixth single-layer motion parameter estimation unit, a second multi-layer motion parameter estimation unit, a seventh single-layer motion parameter estimation unit, and an eighth single-layer motion parameter estimation unit. The fifth, sixth, seventh, and eighth single-layer motion parameter estimation units have the same structure, each including a third normalization layer, a third causal convolutional layer, a second diagonal structured long short-term memory network, a third set of normalization layers, a third projection layer, a second Gaussian error linearization unit, a third residual connection layer, and a third Mamba layer, all connected in sequence. The second multi-layer motion parameter estimation unit includes a fourth normalization layer, a fourth projection layer, a fourth causal convolutional layer, a second neural network combining a diagonal matrix and a long short-term memory network, a fourth set of normalization layers, a second neural network using the Swish activation function, a fourth residual connection layer, and a fourth Mamba layer, all connected in sequence. The backward motion parameter estimation module estimates the first... The vector and the first The features output by the backward motion parameter estimation module are processed to obtain the first... The features output by the backward motion parameter estimation module include: The fifth single-layer motion parameter estimation unit, the sixth single-layer motion parameter estimation unit, the second multi-layer motion parameter estimation unit, the seventh single-layer motion parameter estimation unit, and the eighth single-layer motion parameter estimation unit are cascaded together to estimate the motion parameter estimation of the fifth single-layer motion parameter estimation unit. The vector and the first The features output by the backward motion parameter estimation module are processed to obtain the first... Features output by the backward motion parameter estimation module.

5. The ISAR coherent translational compensation and imaging method based on motion parameter estimation network according to claim 4, characterized in that, The first Mamba, the second Mamba, the third Mamba, and the fourth Mamba have the same structure, each including: a first linear layer, a second linear layer, a convolutional layer, a first swish activation function, a second swish activation function, a state-space model, a third linear layer, and a fully connected layer; the processing procedure of the first Mamba, the second Mamba, the third Mamba, or the fourth Mamba includes: The first linear layer is used to process the input features to obtain the first feature; the first swish activation function is used to process the first feature to obtain the second feature. The input features are processed using the second linear layer to obtain the third feature; the third feature is convolved using the convolutional layer to obtain the fourth feature; and the fourth feature is processed using the second swish activation function to obtain the fifth feature. The state-space model is used to process the fifth feature to obtain the sixth feature; the sixth feature is multiplied by the second feature to obtain the seventh feature; The seventh feature is processed using the third linear layer to obtain the eighth feature; The eighth feature is processed using the fully connected layer to obtain the output feature.

6. The ISAR coherent translational compensation and imaging method based on motion parameter estimation network according to claim 1, characterized in that, The training process of the trained motion parameter estimation network model includes: In the In the next iteration, data similar to the compensated echo sequence is constructed as training data. The initial state of the motion parameter estimation network model to be trained The true distance parameter of the compensated echo sequence is used as the state. ; Get the The action corresponding to the motion parameter estimation network model to be trained; Calculate the initial state With the state The gap loss, according to the first The gap loss, the action, and the learning rate of the training process are backpropagated to update the... The parameters of the motion parameter estimation network model to be trained are obtained. The next training iteration of the motion parameter estimation network model; This process is repeated until the number of training iterations or the degree of convergence meets the preset iteration conditions, at which point the trained motion parameter estimation network model is obtained.

7. The ISAR coherent translational compensation and imaging method based on motion parameter estimation network according to claim 1, characterized in that, Also includes: Based on preset conditions, the preset imaging model reaches the iteration termination condition, the optimal imaging parameters are obtained, and the corresponding imaging results are obtained.

8. The ISAR coherent translational compensation and imaging method based on motion parameter estimation network according to claim 7, characterized in that, The preset condition includes that the difference between the loss function value of the current iteration and the loss function value of the previous iteration is less than a preset value; wherein, the expression for the loss function value of the current iteration is: ； ； in, Represents the image entropy function. Indicates the azimuth length of the image. Indicates the distance length of the image. Indicates the orientation index of the image. Indicates the image distance index. This represents the mean of the imaging results in the current iteration. This represents the imaging result of the current iteration.

9. The ISAR coherent translational compensation and imaging method based on motion parameter estimation network according to claim 1, characterized in that, The step of using the coarse estimate of the translational parameters as the imaging parameters to be optimized in the preset imaging model, and performing iterative optimization to obtain the optimized imaging parameters includes: The compensated echo sequence is used as the imaging parameters to be optimized. Initialize the first-order moments and second moment =0; Calculate the gradient of the loss function value in the current iteration. Its expression is: ； in, Represents the gradient. Represents the image entropy function. This represents the imaging parameters to be optimized in the current iteration. This represents the imaging result of the current iteration. Indicates the orientation index of the image. Indicates the image distance index; The estimated value of the first moment is updated, and its expression is: ； The estimated value of the updated second moment is expressed as follows: ； The updated estimates of the first and second moments are corrected for bias to obtain the corrected estimates of the first moment. and the estimated value of the corrected second moment Their expressions are as follows: ； ； Based on the corrected estimate of the first moment and the estimated value of the corrected second moment The optimized imaging parameter estimates after the current iteration are obtained, and their expression is: ； in, Indicates the learning rate. and These represent the decay rates of the first-order moment estimates and the second-order moment estimates, respectively. Represents a constant.

Images