Space target large-rotation-angle ISAR imaging method based on multi-focus image fusion

Abstract

The invention discloses a space target large-rotation-angle ISAR imaging method based on multi-focus image fusion, and the method comprises the steps: carrying out the space division of a space region where a to-be-measured target is located along the height dimension, obtaining a plurality of subspaces, determining the phase error term of a target echo signal in each subspace, and enabling all subspaces to be not overlapped; the target echo signal is compensated according to the phase error term of each subspace, ISAR image sequences focused in each subspace in a layered mode are obtained, ISAR images correspond to the subspaces in a one-to-one mode, and focusing areas of the ISAR images are in the subspaces corresponding to the ISAR images; inputting the ISAR image sequence into the trained focusing region extraction network to obtain a focusing region of each ISAR image; and performing multi-focus fusion on the ISAR image sequence based on the focus area to obtain a full-focus ISAR image of the target to be detected. The method provided by the invention is good in imaging effect and high in imaging efficiency.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to a spatial target large-angle ISAR imaging method based on multi-focus image fusion. Background Technology

[0002] Inverse Synthetic Aperture Radar (ISAR) can acquire two-dimensional images of targets in all weather conditions and at all times. High-quality, well-focused ISAR images can reveal detailed information such as the size, shape, and structure of targets, which is of great significance for target identification and surveillance tasks. However, during large-angle ISAR imaging of three-dimensional targets, the height coordinates of the scattering point can cause spatially varied phase errors, resulting in image defocusing.

[0003] Traditional spatially varying phase compensation methods mostly model two-dimensional spatially varying phases, which cannot effectively correct image defocus caused by three-dimensional spatially varying phase errors. Although some researchers have proposed a three-dimensional spatially varying phase error compensation method, it suffers from problems such as low phase modeling accuracy, poor adaptability of the segmentation strategy, high algorithm complexity, and long computation time.

[0004] Specifically, this method models the three-dimensional spatially varying phase error as a second-order phase error, while real-world environments exhibit more complex higher-order motions, resulting in limited compensation accuracy. Furthermore, the method employs extensive iterative searches and multiple fractional Fourier transform (FrFT) calculations to solve for the segmented intervals of the echoes, requiring repeated calculations for echoes within each range cell. For large-scale, high-resolution ISAR images, this places a heavy computational burden, making it difficult to meet real-time processing requirements. Additionally, the number of segments depends on the remaining energy threshold parameter setting; too many segments may overfit noise, while too few segments cannot fully compensate for the phase error. There is a lack of reliable methods for adaptively determining the optimal number of segments.

[0005] Therefore, current ISAR imaging methods suffer from poor imaging accuracy and low efficiency. Summary of the Invention

[0006] This invention provides a spatial target large-angle ISAR imaging method based on multi-focus image fusion, which can solve the above-mentioned technical problems.

[0007] In a first aspect, embodiments of the present invention provide a spatial target large-angle ISAR imaging method based on multi-focus image fusion, the method comprising:

[0008] The spatial region where the target is located is divided along the height dimension to obtain multiple subspaces, and the phase error term of the target echo signal in each subspace is determined. All subspaces do not overlap. The target echo signal is compensated according to the phase error term of each subspace to obtain an ISAR image sequence with layered focusing in each subspace, wherein the ISAR image corresponds one-to-one with the subspace, and the focusing area of ​​the ISAR image is within the subspace corresponding to the ISAR image. The ISAR image sequence is input into a trained focus region extraction network to obtain the focus region of each ISAR image; Based on the focused region, the ISAR image sequence is fused using multi-focus fusion to obtain a full-focus ISAR image of the target under test.

[0009] Secondly, embodiments of the present invention provide a spatial target large-angle ISAR imaging device based on multi-focus image fusion, including a spatial division module, a compensation imaging module, a focus region extraction module, and a fusion module; The spatial division module is used to divide the spatial region where the target is located along the height dimension into multiple subspaces, and to determine the phase error term of the target echo signal in each subspace, wherein all subspaces do not overlap. The compensation imaging module is used to compensate the target echo signal according to the phase error term of each subspace, so as to obtain an ISAR image sequence with layered focusing in each subspace, wherein the ISAR image corresponds one-to-one with the subspace, and the focusing area of ​​the ISAR image is within the subspace corresponding to the ISAR image. The focus region extraction module is used to input the ISAR image sequence into the trained focus region extraction network to obtain the focus region of each ISAR image; The fusion module is used to perform multi-focus fusion on the ISAR image sequence based on the focused region to obtain a full-focus ISAR image of the target under test.

[0010] The beneficial effects of this invention compared to the prior art are as follows: By uniformly modeling the three-dimensional spatially varying phase error and accurately stratifying the region where the target is located based on the height dimension, this invention can decouple the three-dimensional spatially varying phase error and improve imaging accuracy; furthermore, the different focal regions of the divided subspaces enable this invention to obtain a fully focused ISAR image through a single image fusion, significantly reducing the computational load during image fusion; simultaneously, this invention extracts the focal region of each ISAR image through a neural network, eliminating the need for iterative search, which greatly reduces the computational complexity of focal region extraction, thereby improving the imaging efficiency of the target. Attached Figure Description

[0011] Figure 1 A schematic diagram of an orbital coordinate system and an imaging coordinate system provided for an embodiment of the present invention; Figure 2 A flowchart illustrating the implementation of a large-angle ISAR imaging method for spatial targets based on multi-focus image fusion, provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a scene where a subspace is divided and layered focusing imaging is performed, as provided in an embodiment of the present invention. Figure 4 This is a flowchart illustrating an image fusion method. Figure 5 This is a schematic diagram of the structure of a focusing region extraction network provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a residual block provided in an embodiment of the present invention; Figure 7 A schematic diagram of an overlapping region provided in an embodiment of the present invention; Figure 8 A schematic diagram comparing a fused image and a fully focused ISAR image provided in an embodiment of the present invention; Figure 9 A flowchart illustrating the implementation of a multi-focus fusion method provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of a space target large-angle ISAR imaging device based on multi-focus image fusion provided in an embodiment of the present invention; Figure 11 A schematic diagram of the three-dimensional scattering point model used in the simulation experiment provided in the embodiments of the present invention; Figure 12 This is a comparative schematic diagram of a fully focused ISAR image provided in an embodiment of the present invention. Detailed Implementation

[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0013] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0014] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0015] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0016] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0017] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0018] 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.

[0019] The space target large-angle ISAR imaging method based on multi-focus image fusion provided in this invention can be applied to electronic devices such as mobile terminals, personal laptops, and supercomputers. This invention does not impose any restrictions on the specific type of electronic device.

[0020] In some embodiments, before imaging, a signal is first transmitted, then the echo signal obtained by the reflected transmitted signal from the target is received, and then range compression and translational compensation are performed on the echo signal to obtain the target echo signal. Subsequent data processing is performed based on the acquired target echo signal.

[0021] In one possible implementation, the space target is typically in a triaxial stable state, with the satellite as the target, and its imaging model is as follows. Figure 1 As shown in (a) of the figure, These represent the initial, center, and final times of a single-frame ISAR imaging. To facilitate imaging analysis, the orbital coordinate system can be transformed to... Figure 1 The equivalent imaging coordinate system shown in (b) is shown in the figure.

[0022] For example, in the equivalent imaging coordinate system, the rotation of the radar line of sight can be equivalent to the spin of a spatial target around a fixed axis, that is, the target being measured rotates around the OZ' axis with an angular velocity Uniform rotation, unit vector of angular velocity azimuth This simplifies imaging analysis.

[0023] Specifically, in the equivalent imaging coordinate system, the OZ' axis rotates along the radar line of sight by a parameter. Direction, along the OX' axis The direction, the OY' axis, and the OX' and OZ' axes satisfy the right-hand rule.

[0024] In one example, the scattering center of the target in the equivalent imaging coordinate system. exist Instantaneous slant distance at time The following formula can be satisfied: (1.1) in, This represents the distance from the reference position of the slant to the origin of the coordinate system. , , )for Coordinates in the equivalent imaging coordinate system The pitch angle, This indicates the image accumulation angle.

[0025] Based on the projection relationship of the k-th frame ISAR image, we can obtain: (1.2) in, These are the projected distance coordinates of the scattering center of the target under test in the azimuth and range directions, respectively.

[0026] If the transmitted signal is a Linear Frequency Modulation (LFM) signal, the target echo signal obtained after range compression and translational compensation is the target echo signal. The following formula can be satisfied: (1.3) in, They represent fast time and slow time, respectively. The scattering intensity at the scattering center of the target is [value]. The pulse width. For pulse accumulation time, The carrier frequency for transmitting signals. For linear modulation frequency, It is the speed of light.

[0027] From the above formula (1.3) As can be seen from the item, in Larger, i.e., under large turning angle conditions, the echo signal is more similar to... The associated three-dimensional spatially variable phase error is not negligible, which is precisely why the traditional planar turntable model imaging method fails.

[0028] Figure 2 The diagram illustrates an implementation flowchart of a large-angle ISAR imaging method for space targets based on multi-focus image fusion, provided by an embodiment of the present invention. This method is illustrative and not limiting, and can be applied to the aforementioned electronic device. The method may include steps S201-S205, which are described below.

[0029] S201, the spatial region where the target to be measured is located is divided along the height dimension to obtain multiple subspaces.

[0030] In one possible implementation, research has shown that when the range migration error in the target echo signal... satisfy , For distance resolution; phase error satisfy At that time, it can be assumed that the corresponding scattering point in the ISAR image is clearly focused. Therefore, based on the derivation of the target echo signal after Fourier transform, it can be concluded that when the target echo signal is subjected to a specific... z After coordinate phase compensation, the phase error of the three-dimensional spatial coding is approximately equal within a certain height range. Therefore, it can be determined according to this height range (i.e., the height where the maximum error is equal). The spatial region where the target is located is divided along the height dimension to obtain multiple subspaces.

[0031] For example, the target echo signal after Fourier transform The following formula can be satisfied: (1.4) in, To adjust the frequency.

[0032] in: (1.5).

[0033] In one example, see Figure 3 It can be extended along the positive and negative directions of the OZ' axis, starting from the scattering center of the target. This yields a subspace whose geometric center coincides with the scattering center of the target object. Then, based on this subspace, it can be expanded along the positive and negative directions of the OZ' axis, respectively. This process is repeated until the top surface of the positive subspace or the bottom surface of the negative subspace exceeds the area where the target is located.

[0034] For example, all subspaces do not overlap.

[0035] In one example, the maximum error equal height can satisfy the following formula: (1.6) in, To ensure the maximum error is equal in height, At the speed of light, The carrier frequency for transmitting signals. The pitch angle, The imaging accumulation angle for a single frame of ISAR image.

[0036] S202, determine the phase error term of the target echo signal in each subspace.

[0037] In one example, within the same subspace, the same phase error term can be used for compensation, and this phase error term can satisfy the following formula: (1.7) in, Let m be the phase error term for the m-th subspace. This is the instantaneous slant distance error. The carrier frequency for transmitting signals. To adjust the frequency.

[0038] For example, referring to the calculation formula (1.7) for the error compensation term, it can be seen that the phase error compensation processes in each subspace are independent of each other. Therefore, MATLAB's parallel computing technology can be used to distribute the compensation operations for phase errors of different height dimensions to different working processes in parallel computing resources.

[0039] S203, the target echo signal is compensated according to the phase error term of each subspace to obtain the ISAR image sequence with layered focusing in each subspace.

[0040] In one example, see Figure 3 It can combine the Polar Format Algorithm (PFA) to correct the remaining two-dimensional spatially variable phase error after the height-dimensional phase compensation in the wavenumber domain, and perform autofocus imaging to generate a layered focused ISAR image sequence.

[0041] For example, in an ISAR image sequence, there is a one-to-one correspondence between ISAR images and subspaces, and the focal region of an ISAR image is within its corresponding subspace.

[0042] S204. Input the ISAR image sequence into the trained focus region extraction network to obtain the focus region of each ISAR image.

[0043] In one example, the loss function used during training of the focus region extraction network can be a weighted sum of the binary cross-entropy loss function, the Dice loss function, and the focus loss function.

[0044] For example, the loss function used by the region extraction network during training. The following formula can be satisfied: (1.8) in, , , These are the binary cross-entropy loss function, the Dice loss function, and the focus loss function, respectively. , , The weights of these three loss functions are listed in order.

[0045] in: (1.9) (1.10) (1.11) in, , Represents events in the true distribution The probability and prediction distribution of events The probability, , These represent the predicted segmentation region and the actual segmentation region, respectively. , These are two regulating factors.

[0046] Optionally, , and These can be set to 0.75 and 2 respectively.

[0047] This invention uses the binary cross-entropy loss function as the basic loss function for classification tasks, which can provide basic directional guidance for prediction. For ISAR images with hierarchical focus, the focused area is usually smaller than the defocused area. This invention uses the Dice loss function to directly evaluate the degree of overlap between the predicted area and the real area, which can better handle the imbalance between the two. By introducing an adjustment factor into the focus loss function, the learning ability of the focus area extraction network for narrow focus areas and focus boundaries can be further enhanced.

[0048] S205, based on the focus region, performs multi-focus fusion on the ISAR image sequence to obtain a full-focus ISAR image of the target under test.

[0049] In one example, theoretically, the focal regions of different ISAR images do not overlap. Therefore, a sub-image within the focal region of each ISAR image can be selected and multi-focus fusion can be performed to obtain a full-focus ISAR image.

[0050] See Figure 4 Traditional image fusion methods, when fusing multiple images, generally employ a pairwise sequential fusion strategy, although the image feature extraction and fusion operations may vary slightly depending on the algorithm design. Specifically, image 1 and image 2 are selected as the images to be fused. Two focus region evaluation maps are generated using a feature extraction network and a focus evaluation algorithm, respectively. A decision map is obtained through two-channel max pooling. Then, image 1 and image 2 are fused based on the decision map to obtain fused image 1. Next, fused image 1 and image 3 are selected as the images to be fused, and a decision map is obtained to obtain fused image 2. This process is repeated until all images to be fused are fused, resulting in the final fused image. This method requires 2(N) steps when the number of images is N. 1) Sub-feature extraction process. Since the present invention performs layered imaging based on subspace during imaging, the focus areas of different ISAR regions do not overlap. Therefore, the focus areas of all ISAR images can be directly extracted and fused in one step to obtain a fully focused ISAR image, which can significantly reduce the amount of computation and improve the fusion efficiency.

[0051] This invention achieves decoupling of three-dimensional spatially varying phase errors and improves imaging accuracy by uniformly modeling the three-dimensional spatially varying phase error and accurately stratifying the region where the target is located based on the height dimension. Furthermore, the different focal regions of the divided subspaces enable this invention to obtain a fully focused ISAR image through a single image fusion, significantly reducing the computational load during image fusion. Simultaneously, this invention extracts the focal region of each ISAR image through a neural network, eliminating the need for iterative search and greatly reducing the computational complexity of focal region extraction, thereby improving the imaging efficiency of the target.

[0052] Example 2 Figure 5 The diagram shown illustrates the structure of a focused region extraction network according to an embodiment of the present invention. As an example and not a limitation, the network may include a basic feature extraction module, a deep feature extraction module, a feature integration and compression module, and an output layer.

[0053] In one possible implementation, see Figure 5 A three-channel ISAR image of size H×W×3 can be input into the basic feature extraction module. This module uses a 3×3 convolutional kernel to increase the input channels from 3 to 64, thereby extracting shallow features such as edges and textures. Next, a deep feature extraction module, composed of multiple cascaded residual blocks, gradually abstracts more complex and global semantic features from the low-level shallow features, obtaining deep features, which are crucial for accurately distinguishing between focused and defocused areas. The feature integration and compression module integrates and compresses the deep features, and finally, the output layer determines the focused region of the ISAR image.

[0054] In one example, see Figure 6 The structure and data processing flow of the residual blocks are shown. The stacked residual blocks provide a "highway" to the early layers for the deep focus region extraction network, thus solving the gradient vanishing problem in the training process of deep networks. At the same time, shallow features are transmitted through dense connections, which avoids the network relearning redundant features, allowing the network to focus more on fine-tuning and enhancing existing features.

[0055] In one example, the feature integration and compression module can use a 3×3 convolutional layer for further feature integration, preparing for the final output. The output layer can use a 3×3 convolutional layer to compress the 64-channel feature map into a single channel, and use the Sigmoid function to map the scalar value of each pixel to the [0,1] interval, thereby obtaining the focus region corresponding to the ISAR image.

[0056] The focus region extraction network provided in this invention, by combining a residual learning mechanism and multi-level feature extraction, can effectively model the focus and defocus regions in ISAR images, thereby obtaining the focus region image. The residual learning mechanism, through dense skip connections, can solve the gradient vanishing and gradient exploding problems during the training process of deep neural networks. The focus region extraction network adopts a fully convolutional architecture, eliminating the need for fully connected layers, supporting ISAR image inputs of arbitrary sizes while maintaining the resolution of the output focus region image.

[0057] Example 3 Ideally, the focus region extraction network extracts the actual focus regions of each ISAR image, with no overlap between them. A full-focus ISAR image of the target can be obtained through a single image fusion. However, see... Figure 7 To ensure accuracy, the focus region extracted by the focus region extraction network is often slightly larger than the actual focus region of each ISAR image. Generally, the overlap between two adjacent focus regions is usually narrow and located at the boundary of the focus region, resulting in extremely weak defocusing and no obvious defocus diffusion; however, see... Figure 8 In (a), the repeated accumulation of energy leads to overflow of scattering point energy exceeding 255 in the overlapping area, resulting in lower brightness in the corresponding image area compared to normal scattering points. This causes differences in brightness across different regions, leading to significant differences in scattering point energy between different regions in the fused image and poor image quality. Therefore, correction is required using the following method to obtain the final full-focus ISAR image.

[0058] Figure 9 The diagram shown illustrates a multi-focus fusion method according to an embodiment of the present invention. The method may include steps S901-S904, which are described below.

[0059] S901: Select a sub-image within the focal region of each ISAR image, perform multi-focus fusion, and obtain a fused image.

[0060] In one example, the fused images can satisfy the following formula: (1.12) in, To merge images, For the first ISAR images The focal region, where N is the number of ISAR images.

[0061] S902, determine the overlapping areas between the focus areas.

[0062] In one example, the overlap area map (Omap) can be determined using the following formula: (1.13) in, For the first ISAR images The focal region, where N is the number of ISAR images. This indicates binarization processing.

[0063] S903, determine the weights of the fused image based on the overlapping region.

[0064] (1.14) in, To fuse the image weights, This is a mask for the overlapping area.

[0065] S904, the fused image is corrected according to the weights to obtain a full-focus ISAR image.

[0066] In one example, a fully focused ISAR image can be determined using the following formula: (1.15) in, For full-focus ISAR images, To fuse the image weights, To merge images.

[0067] For example, see Figure 8 In (a), the overlapping area of ​​the two images consists of both an overlapping area of ​​focused region (OF) and an overlapping area of ​​defocused region (OD). Assuming that both contribute equally to the overlapping area, the fusion result after correction of the overlapping area of ​​the two images is calculated according to the above formulas (1.12)~(1.15). For visual effects, please refer to... Figure 8 As shown in (b), by correcting the fused image, the OF and OD regions in the image to be fused (the predicted focus region sub-image of the ISAR image) can be weighted and averaged to weaken the effect of defocus diffusion and solve the problem of pixel value overflow caused by repeated accumulation of energy of scattering points in the fused image.

[0068] Therefore, the multi-focus fusion method provided by this invention can weaken the influence of defocus diffusion effect through simple weighting, and at the same time solve the problem of pixel value overflow caused by repeated accumulation of energy of scattering points in the fused image.

[0069] Example 5 Figure 10 The diagram shown is a schematic representation of a large-angle ISAR imaging device for space targets based on multi-focus image fusion, provided by an embodiment of the present invention. As an example and not a limitation, the device may include a spatial segmentation module 110, a compensation imaging module 120, a focus region extraction module 130, and a fusion module 140.

[0070] For example, the spatial partitioning module 110 is used to spatially partition the spatial region where the target to be measured is located along the height dimension to obtain multiple subspaces, and determine the phase error term of the target echo signal in each subspace, wherein all subspaces do not overlap; the compensation imaging module 120 is used to compensate the target echo signal according to the phase error term of each subspace to obtain an ISAR image sequence with layered focusing in each subspace, wherein the ISAR image corresponds one-to-one with the subspace, and the focusing region of the ISAR image is within the subspace corresponding to the ISAR image; the focusing region extraction module 130 is used to input the ISAR image sequence into a trained focusing region extraction network to obtain the focusing region of each ISAR image; the fusion module 140 is used to perform multi-focus fusion on the ISAR image sequence based on the focusing region to obtain a full-focus ISAR image of the target to be measured.

[0071] This invention achieves decoupling of three-dimensional spatially varying phase errors and improves imaging accuracy by uniformly modeling the three-dimensional spatially varying phase error and accurately stratifying the region where the target is located based on the height dimension. Furthermore, the different focal regions of the divided subspaces enable this invention to obtain a fully focused ISAR image through a single image fusion, significantly reducing the computational load during image fusion. Simultaneously, this invention extracts the focal region of each ISAR image through a neural network, eliminating the need for iterative search and greatly reducing the computational complexity of focal region extraction, thereby improving the imaging efficiency of the target.

[0072] To better illustrate the beneficial effects of the present invention, the following simulation experiments were conducted: For example, in the simulation experiment, a self-checking three-dimensional scattering point model was selected, and the parameters of the model were modeled after, for example, Figure 11 The satellite is shown. To verify the effectiveness of the proposed method, the target echo is subjected to hierarchical compensation for three-dimensional spatially varied phase error and multi-focus fusion to generate a full-focus ISAR image. Furthermore, by removing the spatially varied phase error during modeling, an ideal full-focus ISAR image can be obtained. The similarity between the acquired full-focus ISAR image and the ideal result is evaluated to verify the focusing effect of the full-focus ISAR image and the accuracy of the three-dimensional spatially varied phase error compensation.

[0073] Specifically, in the simulation experiment, echoes were obtained using a self-built scattering point model under different attitudes, and layered compensation for three-dimensional spatially varying phase errors was performed on the echoes. The Polar Format Algorithm (PFA) was used to correct the range migration of the echoes, and then a 512×512 pixel layered focused ISAR image sequence was generated using the RD imaging algorithm. The image sequences were grouped as datasets for the ResNet network (i.e., the focused region extraction network), and the details of the dataset grouping are shown in Table 1. The initial learning rate of the ResNet network was 7e.-5 The training frequency was reduced by a factor of 0.7 after the 30th, 60th, and 80th epochs, with 100 training rounds. The trained network was used to extract the focus regions of the test set, and a parallel fusion strategy was employed to fuse the focus regions. Finally, the overlapping region processing module was used to generate a full-focus ISAR image.

[0074] Table 1

[0075] Figure 12 The diagram shown is a comparative illustration of a fully focused ISAR image provided in an embodiment of the present invention.

[0076] See Figure 12 , Figure 12 Column (a) in the text represents a single layered focused image, which is the ISAR image sequence mentioned in this invention. Figure 12 Column (b) in the image represents an ideal full-focus ISAR image. Figure 12 (c) in the figure is a fully focused ISAR image obtained by the method provided in this invention.

[0077] As can be seen, the method provided by this invention can achieve layered compensation for three-dimensional spatially varying phase errors and generate refocused ISAR images with focusing effects very close to ideal fully focused ISAR images.

[0078] In addition, to quantify the imaging effect of the present invention, seven standard similarity evaluation metrics were used for evaluation: Mean Square Error (MSE), Root Mean Square Error (RMSE), Peak Signal to Noise Ratio (PSNR), Structural Similarity (SSIM), General Quality Index (UQI), Multi-scale Structural Similarity (MS_SSIM), and Spatial Correlation Coefficient (SCC).

[0079] Specifically, the above evaluation indicators can be calculated using the following formulas: (1.16) (1.17) (1.18) (1.19) (1.20) (1.21) (1.22) in, , Let I and K be constants, and let I and K be the full-focus ISAR image and the ideal full-focus ISAR image, respectively. The mean, For variance, For covariance, These are the brightness, contrast, and structural contrast functions, respectively.

[0080] Referring to the specific values ​​of the evaluation indicators shown in Table 2 below, it can be seen that the seven standard similarity evaluation indicators of the full-focus ISAR image obtained by the method provided by the present invention are all better than the empirical values ​​and very close to the ideal evaluation indicators. Moreover, the PSNR indicator reaches an excellent level, proving that the full-focus ISAR image generated by the present invention is clearly focused and of high quality. It also verifies the accuracy of the three-dimensional spatially varied phase error compensation method provided by the present invention.

[0081] Table 2

[0082] Therefore, this invention achieves decoupling of three-dimensional spatially varying phase errors and improves imaging accuracy by uniformly modeling the three-dimensional spatially varying phase error and accurately stratifying the region where the target is located based on the height dimension. Furthermore, the different focal regions of the divided subspaces enable this invention to obtain a fully focused ISAR image through a single image fusion, significantly reducing the computational load during image fusion. At the same time, this invention extracts the focal region of each ISAR image through a neural network, eliminating the need for iterative search and greatly reducing the computational complexity of focal region extraction, thereby improving the imaging efficiency of the target.

[0083] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

Claims

1. A method for large-angle ISAR imaging of space targets based on multi-focus image fusion, characterized in that, include: The spatial region where the target is located is divided along the height dimension to obtain multiple subspaces, and the phase error term of the target echo signal in each subspace is determined. All subspaces do not overlap. The target echo signal is compensated according to the phase error term of each subspace to obtain an ISAR image sequence with layered focusing in each subspace, wherein the ISAR image corresponds one-to-one with the subspace, and the focusing area of ​​the ISAR image is within the subspace corresponding to the ISAR image. The ISAR image sequence is input into a trained focus region extraction network to obtain the focus region of each ISAR image; Based on the focused region, the ISAR image sequence is fused using multi-focus fusion to obtain a full-focus ISAR image of the target under test.

2. The method according to claim 1, characterized in that, The height of each subspace is the same as the maximum error height. The three-dimensional spatial phase error of each point in the same subspace is approximately equal. There exists a subspace whose geometric center coincides with the scattering center of the target under test.

3. The method according to claim 2, characterized in that, The maximum error equal height satisfies the following formula: in, To ensure the maximum error is equal in height, At the speed of light, The carrier frequency for transmitting signals. The pitch angle, The imaging accumulation angle for a single frame of ISAR image.

4. The method according to claim 3, characterized in that, The phase error term satisfies the following formula: in, Let m be the phase error term for the m-th subspace. This is the instantaneous slant distance error. The carrier frequency for transmitting signals. To adjust the frequency.

5. The method according to claim 1, characterized in that, The focused region extraction network includes a basic feature extraction module, a deep feature extraction module, a feature integration and compression module, and an output layer; The basic feature extraction module is used to extract shallow features from the ISAR image, and the deep feature extraction module is used to extract deep features from the basic features of the ISAR image. The feature integration and compression module is used to integrate and compress the deep features, and the output layer is used to determine the focus area of ​​the ISAR image based on the integrated and compressed features. The deep feature extraction module is composed of multiple cascaded residual blocks, and the loss function used by the focus region extraction network during training is a weighted sum of the binary cross-entropy loss function, the Dice loss function, and the focus loss function.

6. The method according to claim 1, characterized in that, The step of performing multi-focus fusion on the ISAR image sequence based on the focused region to obtain a full-focus ISAR image of the target includes: Sub-images within the focal region of each ISAR image are selected and multi-focus fusion is performed to obtain a fused image; Identify the overlapping regions between the focused regions, and determine the weights of the fused image based on the overlapping regions; The fused image is corrected according to the weights to obtain the fully focused ISAR image.

7. The method according to claim 6, characterized in that, The fused image satisfies the following formula: in, For the fused image, For the first ISAR images The focal region, where N is the number of ISAR images.

8. The method according to claim 6, characterized in that, The weights of the fused image satisfy the following formula: in, The weights of the fused image, A mask for overlapping regions; in: in, For the first ISAR images The focal region, where N is the number of ISAR images. This indicates binarization processing.

9. The method according to claim 6, characterized in that, The fully focused ISAR image satisfies the following formula: in, The full-focus ISAR image, The weights of the fused image, The fused image.

10. A space target large-angle ISAR imaging device based on multi-focus image fusion, characterized in that, It includes a spatial segmentation module, a compensation imaging module, a focus area extraction module, and a fusion module; The spatial division module is used to divide the spatial region where the target is located along the height dimension into multiple subspaces, and to determine the phase error term of the target echo signal in each subspace, wherein all subspaces do not overlap. The compensation imaging module is used to compensate the target echo signal according to the phase error term of each subspace, so as to obtain an ISAR image sequence with layered focusing in each subspace, wherein the ISAR image corresponds one-to-one with the subspace, and the focusing area of ​​the ISAR image is within the subspace corresponding to the ISAR image. The focus region extraction module is used to input the ISAR image sequence into the trained focus region extraction network to obtain the focus region of each ISAR image; The fusion module is used to perform multi-focus fusion on the ISAR image sequence based on the focused region to obtain a full-focus ISAR image of the target under test.

Images