Method and apparatus for target detection in SAR azimuth-ambiguous images

Abstract

This disclosure provides a method, apparatus, device, and storage medium for target detection in SAR azimuth-ambiguous interference images, which can be applied to the fields of SAR signal processing and intelligent interpretation technology of synthetic aperture radar. The method includes: performing azimuth-to-Fourier transform processing on 1A-level complex data containing azimuth ambiguity to obtain range-Doppler domain data; inputting the range-Doppler domain data into a spectral estimation model to obtain a spectral estimation map, wherein the spectral estimation model is trained based on a simulated spectrum dataset constructed using random parameters, and the spectral estimation model includes a multi-scale feature extraction module and a spectral sequence analysis module; extracting sub-spectral segments based on the spectral estimation map and sub-spectral proportions; extrapolating the sub-spectral segments to obtain image data with azimuth ambiguity suppressed; and processing the image data with suppressed azimuth ambiguity before inputting it into a target detection model to obtain the final detection result.

Description

Technical Field

[0001] This disclosure relates to the field of SAR signal processing and intelligent interpretation technology for synthetic aperture radar, and in particular to a target detection method, apparatus, device, medium and program product for SAR azimuth-ambiguous interference images. Background Technology

[0002] Spaceborne SAR (Synthetic Aperture Radar) is an active imaging system capable of performing imaging tasks despite limitations imposed by conditions such as clouds, rain, fog, and darkness. With its all-weather, all-time imaging capabilities, spaceborne SAR offers unique advantages in target detection, target change monitoring, and target tracking, and is widely used in military, environmental monitoring, and resource exploration fields.

[0003] Synthetic aperture radar is highly susceptible to azimuth ambiguity when imaging large dynamic scenes such as land-sea boundaries and water surfaces, resulting in false targets in the image. This problem can easily lead to false detections and missed detections in target detection tasks.

[0004] However, in related technologies, the azimuth ambiguity suppression method based on the concept of selective filters is only applicable to strip operation mode and requires accurate imaging parameters to construct the selective filter. It is difficult to use properly under harsh conditions where parameters are missing or inaccurate. Meanwhile, intelligent interpretation methods for SAR images do not consider the interference caused by azimuth ambiguity, leading to false positives and false negatives when interpreting images affected by azimuth ambiguity. Summary of the Invention

[0005] In view of the above problems, this disclosure provides a method, apparatus, device, medium and program product for target detection in SAR azimuth-ambiguous interference images.

[0006] According to a first aspect of this disclosure, a target detection method for SAR azimuth-ambiguous interference images is provided, comprising:

[0007] Azimuth-to-Fourier transform is performed on the 1A-level complex data containing azimuth ambiguity to obtain range-Doppler domain data. The aforementioned 1A-level complex data containing azimuth ambiguity represents the original image data affected by azimuth ambiguity.

[0008] The above-mentioned range Doppler domain data is input into the spectral estimation model to obtain the spectral estimation map. The spectral estimation model is trained based on the simulated spectrum dataset, which is constructed using random parameters. The simulated spectrum dataset includes the simulated spectrum and the interference map corresponding to the simulated spectrum. The spectral estimation model includes a multi-scale feature extraction module and a spectral sequence analysis module.

[0009] Based on the above spectrum estimation diagram and sub-spectrum proportion, sub-spectrum segments are extracted, where the above sub-spectrum proportion represents the ratio of the frequency band in which the above sub-spectrum segment is located to the pulse repetition period;

[0010] Extrapolating the above sub-spectral segments yields image data with suppressed azimuth blur, wherein the image with suppressed azimuth blur has the same azimuth resolution as the original image.

[0011] The image data with suppressed orientation blur is processed and then input into the target detection model to obtain the final detection result.

[0012] According to embodiments of this disclosure, the above-mentioned simulated spectrum dataset is constructed using random parameters and includes:

[0013] The final amplitude sequence in the range direction is obtained based on the inverse Gaussian distribution random amplitude sequence along the range direction and the point spread function, wherein the aforementioned inverse Gaussian distribution random amplitude sequence along the range direction is generated based on the aforementioned random parameters.

[0014] Based on the final amplitude sequence of the range direction and the migration line equation corresponding to the final amplitude sequence of the range direction, the two-dimensional amplitude spectrum is obtained;

[0015] The above two-dimensional amplitude spectrum is subjected to azimuth antenna pattern weighting to obtain fuzzy discrimination and principal discrimination.

[0016] The above-mentioned fuzzy distinguishing factor and the above-mentioned principal distinguishing factor are superimposed and coherent noise is added to obtain the above-mentioned simulated spectrum.

[0017] According to embodiments of this disclosure, the target detection method for SAR azimuth-ambiguous interference images further includes:

[0018] The above two-dimensional amplitude spectra are superimposed to obtain an interference map corresponding to the simulated spectrum; and

[0019] The above-mentioned spectrum estimation model is trained based on the above-mentioned simulated spectrum and the above-mentioned interference diagram corresponding to the above-mentioned simulated spectrum, and the trained spectrum estimation model is obtained.

[0020] According to embodiments of this disclosure, the multi-scale feature extraction module includes a first multi-scale feature extraction module and a second multi-scale feature extraction module. The step of inputting the range-Doppler domain data into the spectral estimation model to obtain the spectral estimation map includes:

[0021] The above-mentioned range Doppler domain data is input into the first multi-scale feature extraction module in the above-mentioned spectral estimation model to obtain feature maps of different sizes;

[0022] The feature maps of the different sizes mentioned above are decomposed to obtain the first sequence;

[0023] The first sequence is input into the spectral sequence analysis module to obtain the second sequence;

[0024] The second sequence described above is reconstructed using feature maps to obtain a reconstructed feature map; and

[0025] The reconstructed feature map is input into the second multi-scale feature extraction module to obtain the spectral estimation map.

[0026] According to embodiments of this disclosure, the above-mentioned extraction of subspectral segments based on the above-mentioned spectral estimation diagram and subspectral proportions includes:

[0027] Averaging the above spectral estimation along the range direction yields a one-dimensional vector ordered by azimuth frequency.

[0028] Based on the aforementioned one-dimensional vector and the aforementioned subspectral proportions, the aforementioned subspectral segments are determined; and

[0029] Based on the above-mentioned Class 1A complex data with orientation ambiguity and the selected Fourier transform matrix, the above-mentioned sub-spectral segments are extracted.

[0030] According to embodiments of this disclosure, the extrapolation of the aforementioned subspectral bands to obtain the aforementioned image data with suppressed orientation blur includes:

[0031] Based on the above sub-spectral segments and extrapolation formula, the extrapolation result is obtained; and

[0032] The above extrapolation results are subjected to an inverse Fourier transform of orientation to obtain image data with the orientation blur suppressed.

[0033] According to embodiments of this disclosure, the image data with suppressed orientation blur is processed and then input into the target detection model to obtain the final detection result, including:

[0034] The image data with suppressed orientation blur is subjected to amplitude plotting and truncated quantization; and

[0035] The processed image data is input into the target detection model to obtain the final detection result.

[0036] The second aspect of this disclosure provides a target detection device for SAR azimuth-ambiguous interference images, comprising: a first acquisition module, a second acquisition module, a cropping module, a third acquisition module, and a fourth acquisition module. The first acquisition module performs an azimuth-to-Fourier transform on azimuth-ambiguous 1A-level complex data to obtain range-Doppler domain data, wherein the azimuth-ambiguous 1A-level complex data represents the original image data affected by azimuth-ambiguity interference. The second acquisition module inputs the range-Doppler domain data into a spectral estimation model to obtain a spectral estimation map, wherein the spectral estimation model is trained based on a simulated spectrum dataset constructed using random parameters, the simulated spectrum dataset including a simulated spectrum and an interference map corresponding to the simulated spectrum, and the spectral estimation model including a multi-scale feature extraction module and a spectral sequence analysis module. The cropping module crops sub-spectral segments based on the spectral estimation map and sub-spectral proportions, wherein the sub-spectral proportion represents the ratio of the frequency band of the sub-spectral segment to the pulse repetition period. The third acquisition module is used to extrapolate the aforementioned sub-spectral segments to obtain image data with suppressed azimuth blur, wherein the image with suppressed azimuth blur has the same azimuth resolution as the original image. The fourth acquisition module is used to process the image data with suppressed azimuth blur and input it into the target detection model to obtain the final detection result.

[0037] A third aspect of this disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.

[0038] A fourth aspect of this disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the methods described above.

[0039] The fifth aspect of this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0040] According to the target detection method, apparatus, device, medium, and program products for SAR azimuth-ambiguous interference images provided in this disclosure, azimuth-to-Fourier transform processing is performed on 1A-level complex data containing azimuth ambiguity, and the obtained range-Doppler domain data is input into a spectral estimation model to obtain a spectral estimation map. The spectral estimation model is trained using a simulated spectrum dataset constructed with random parameters. The construction of the simulated spectrum dataset introduces random parameters, enabling the spectral estimation model to adaptively process SAR images with different imaging parameters even with unknown parameters. Extrapolation of subspectral segments obtained from the spectral estimation map and subspectral proportions yields image data with suppressed azimuth ambiguity. This method effectively suppresses azimuth ambiguity in SAR images acquired in all working modes and does not require imaging parameters, allowing it to be used under adverse conditions such as missing or inaccurate parameters. Finally, during target detection, the processed image data with suppressed azimuth ambiguity is input into the target detection model, resolving the problem of false detections and missed detections caused by azimuth ambiguity, thus obtaining the final detection result. Attached Figure Description

[0041] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0042] Figure 1 The illustration schematically depicts an application scenario of a target detection method, apparatus, device, medium, and program product for SAR azimuth-ambiguous interference images according to embodiments of the present disclosure;

[0043] Figure 2 A flowchart illustrating a target detection method for SAR azimuth-ambiguous interference images according to an embodiment of the present disclosure is shown schematically.

[0044] Figure 3 A flowchart illustrating the construction of a simulated spectrum dataset according to an embodiment of the present disclosure is shown schematically;

[0045] Figure 4 The illustration schematically shows a process flow diagram for constructing a simulated spectrum dataset for training a spectrum estimation model according to an embodiment of the present disclosure;

[0046] Figure 5 A flowchart illustrating the process of obtaining a spectral estimation plot according to an embodiment of the present disclosure is shown schematically;

[0047] Figure 6 A schematic diagram illustrating the structure of a spectral estimation model according to an embodiment of the present disclosure is shown.

[0048] Figure 7 A flowchart illustrating the extraction of sub-spectral segments according to an embodiment of the present disclosure is shown schematically;

[0049] Figure 8 A schematic diagram illustrating the structure of a target detection method for SAR azimuth-ambiguous interference images according to an embodiment of the present disclosure is shown.

[0050] Figure 9 schematically illustrates a comparative experiment using satellite remote sensing data according to an embodiment of the present disclosure;

[0051] Figure 10 schematically shows the effect of the orientation blur suppression method of the present disclosure in a comparative experiment according to an embodiment of the present disclosure;

[0052] Figure 11 schematically illustrates the effect after processing by a conventional adaptive micro / nano filter method according to an embodiment of the present disclosure;

[0053] Figure 12 schematically illustrates a real test data sample extracted from a test dataset for target detection testing according to an embodiment of the present disclosure;

[0054] Figure 13 A schematic block diagram of a target detection apparatus for SAR azimuth-blurred images according to embodiments of the present disclosure is shown; and

[0055] Figure 14 A block diagram of an electronic device suitable for implementing a target detection method for SAR azimuth-blurred images is illustrated according to an embodiment of the present disclosure. Detailed Implementation

[0056] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0057] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0058] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0059] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0060] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of user personal information comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and there is no violation of public order and good morals.

[0061] In the technical solution disclosed herein, the acquisition, collection, storage, use, processing, transmission, provision, disclosure, and application of data all comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.

[0062] In implementing this disclosure, it was discovered that when the PRF (Pulse Repetition Frequency) is low and insufficient to concentrate most of the signal energy within a single PRF period, discrete sampling in the azimuth direction can cause fuzzy energy to alias into a single PRF period, resulting in azimuth ambiguity. In related technologies, there are various azimuth ambiguity suppression methods based on the concept of selective filters; however, they are only applicable to stripe operation modes and require accurate imaging parameters to construct the selective filters, making them difficult to use properly under harsh conditions of missing or inaccurate parameters. Meanwhile, for intelligent interpretation of SAR images, most accurate and efficient methods are based on deep learning algorithms. They utilize convolutional neural network models with varying structures to quickly and accurately complete target detection tasks. However, the problem with these methods is that they do not consider the interference caused by azimuth ambiguity, leading to false detections and missed detections when interpreting images affected by azimuth ambiguity.

[0063] Therefore, embodiments of this disclosure provide a target detection method for SAR azimuth-ambiguous interference images, comprising: performing azimuth-directed Fourier transform processing on azimuth-ambiguous 1A-level complex data to obtain range-Doppler domain data, wherein the azimuth-ambiguous 1A-level complex data represents the original image data affected by azimuth-ambiguity interference; inputting the range-Doppler domain data into a spectral estimation model to obtain a spectral estimation map, wherein the spectral estimation model is trained based on a simulated spectrum dataset, the simulated spectrum dataset is constructed using random parameters, the simulated spectrum dataset includes a simulated spectrum and an interference map corresponding to the simulated spectrum, and the spectral estimation model includes a multi-scale feature extraction module and a spectral sequence analysis module; extracting sub-spectral segments based on the spectral estimation map and sub-spectral proportions, wherein the sub-spectral proportion represents the ratio of the frequency band of the sub-spectral segment to the pulse repetition period; extrapolating the sub-spectral segments to obtain azimuth-ambiguity-suppressed image data, wherein the azimuth-ambiguity-suppressed image has the same azimuth resolution as the original image; and processing the azimuth-ambiguity-suppressed image data and inputting it into a target detection model to obtain the final detection result.

[0064] Figure 1 The illustration shows an application scenario diagram for target detection in SAR azimuth-ambiguous interference images according to embodiments of the present disclosure.

[0065] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0066] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0067] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0068] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0069] For example, server 105 can perform azimuth-to-Fourier transform on the 1A-level complex data containing azimuth ambiguity to obtain range-Doppler domain data. The range-Doppler domain data is then input into the spectral estimation model to obtain a spectral estimation map. Based on the spectral estimation map and the proportion of sub-spectral segments, sub-spectral segments are extracted and extrapolated to obtain image data with suppressed azimuth ambiguity. Finally, the image data with suppressed azimuth ambiguity is processed and input into the target detection model to obtain the final detection result.

[0070] It should be noted that the target detection method for SAR azimuth-blurred interference images provided in this disclosure embodiment can generally be executed by server 105. Correspondingly, the target detection device for SAR azimuth-blurred interference images provided in this disclosure embodiment can generally be located in server 105. The target detection method for SAR azimuth-blurred interference images provided in this disclosure embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the target detection device for SAR azimuth-blurred interference images provided in this disclosure embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0071] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0072] The following will be based on Figure 1 The described scene, through Figures 2-1 2. A detailed description of the target detection method for SAR azimuth-ambiguous interference images according to the disclosed embodiments is provided.

[0073] Figure 2 A flowchart illustrating a target detection method for SAR azimuth-ambiguous interference images according to an embodiment of the present disclosure is shown schematically.

[0074] like Figure 2As shown, the method 200 includes operations S210 to S250.

[0075] In operation S210, the azimuth-to-Fourier transform is performed on the 1A-level complex data containing azimuth ambiguity to obtain range-Doppler domain data.

[0076] According to embodiments of this disclosure, Class 1A complex data containing azimuth ambiguity can characterize the original image data affected by azimuth ambiguity interference. Class 1A complex data containing azimuth ambiguity can be output using a SAR ground receiving system.

[0077] According to embodiments of this disclosure, 1A-level complex data can characterize image data that, after imaging and radiometric correction, retains amplitude and phase information and is stored in complex form. The specific algorithm for applying the azimuth Fourier transform can be represented as a Fast Fourier Transform.

[0078] According to embodiments of this disclosure, Class 1A complex data Y containing orientation ambiguity T And the corresponding distance-Doppler domain data Y D Both can characterize N a Line N r A complex matrix of columns, where N a N can represent the initial number of azimuth sampling points. r It can represent the initial distance to the number of sampling points.

[0079] According to embodiments of this disclosure, the purpose of performing an azimuth Fourier transform on 1A-level complex data is to obtain the corresponding range-Doppler domain data, i.e., the range-Doppler domain spectrum. 1A-level complex data can represent two-dimensional time-domain data, while range-Doppler domain data can represent frequency-domain data. For example, in a two-dimensional coordinate system, the x-axis can represent the range direction, and the y-axis can represent the azimuth direction. Performing an azimuth Fourier transform on 1A-level complex data can represent performing a Fourier transform on one direction of the 1A-level complex data, which is equivalent to performing a Fourier transform on each label sequence in one direction of the two-dimensional data to obtain frequency-domain data, i.e., range-Doppler domain data.

[0080] In operation S220, the distance Doppler domain data is input into the spectral estimation model to obtain the spectral estimation map.

[0081] According to embodiments of this disclosure, the spectral estimation model is trained on a simulated spectrum dataset, which is constructed using random parameters. The simulated spectrum dataset may include a simulated spectrum and an interference map corresponding to the simulated spectrum. The spectral estimation model may include a multi-scale feature extraction module and a spectral sequence analysis module.

[0082] According to embodiments of this disclosure, the distance Doppler domain data Y is... DInputting the spectral estimation model M, we obtain the spectral estimation map E of the range-Doppler domain data. D Spectral estimation diagram E D N can be represented a Line N r A columnar real matrix, where each sample point in the real matrix is ​​a decimal between 0 and 1, can represent the distance to the Doppler domain data Y. D The ratio of the fuzzy signal intensity at the corresponding sample point to the overall signal intensity is proportional to the degree of fuzzy energy interference at that sample point. Therefore, the spectral estimation diagram E... D Each sample point in the data corresponds to a distance from the Doppler domain data Y. D The degree of interference from fuzzy energy at each sample point.

[0083] When operating S230, sub-spectral segments are extracted based on the spectral estimation diagram and the proportion of sub-spectral segments.

[0084] According to embodiments of this disclosure, the subspectral proportion can characterize the ratio of the frequency band containing the subspectral segment to the pulse repetition period. The subspectral segment can characterize the spectral segment least affected by azimuth ambiguity energy interference.

[0085] In operation S240, the sub-spectral bands are extrapolated to obtain image data with suppressed azimuth ambiguity.

[0086] According to embodiments of this disclosure, the image with suppressed azimuth blur has the same azimuth resolution as the original image. For the retained subspectral bands... Extrapolation yields the extrapolated result Y. E extrapolation result Y E Performing an inverse azimuth-to-Fourier transform can yield image data Y with azimuth blur suppressed. R The specific algorithm for applying the azimuth-directed inverse Fourier transform can be represented as the inverse fast Fourier transform.

[0087] According to embodiments of this disclosure, the purpose of extrapolation is to ensure that the final image with suppressed azimuth blur has the same azimuth resolution as the original image. The extrapolation result Y... E The purpose of performing inverse azimuth to Fourier transform is to obtain temporal data with suppressed azimuth blur, that is, image data with suppressed azimuth blur.

[0088] In operation S250, the image data with suppressed orientation blur is processed and then input into the target detection model to obtain the final detection result.

[0089] According to embodiments of this disclosure, azimuth-to-Fourier transform processing is performed on 1A-level complex data containing azimuth ambiguity, and the resulting range-Doppler domain data is input into a spectral estimation model to obtain a spectral estimation map. The spectral estimation model is trained using a simulated spectrum dataset constructed with random parameters. The construction of the simulated spectrum dataset introduces random parameters, enabling the spectral estimation model to adaptively process SAR images with different imaging parameters even with unknown parameters. Extrapolation of subspectral segments obtained from the spectral estimation map and subspectral proportions yields image data with suppressed azimuth ambiguity. This method effectively suppresses azimuth ambiguity in SAR images acquired in all operating modes and does not require imaging parameters, allowing it to be used under adverse conditions such as missing or inaccurate parameters. Finally, during target detection, the processed image data with suppressed azimuth ambiguity is input into the target detection model, resolving the problem of false detections and missed detections caused by false targets resulting from azimuth ambiguity, thus obtaining the final detection result.

[0090] Figure 3 A flowchart illustrating the construction of a simulated spectrum dataset according to an embodiment of this disclosure is shown.

[0091] like Figure 3 As shown, the method 300 includes operations S310 to S340.

[0092] In operation S310, the final amplitude sequence in the range direction is obtained based on the inverse Gaussian distribution random amplitude sequence along the range direction and the corresponding point spread function.

[0093] According to embodiments of this disclosure, the inverse Gaussian distributed random amplitude sequence along the range direction is generated based on random parameters, wherein the range of the random parameters is determined based on the spaceborne radar and the real-world scenario. A′ -1 A′ +1 A and A′0 can represent amplitude sequences randomly generated according to an inverse Gaussian distribution, respectively, which simulate the amplitude distribution of general image data along the range direction.

[0094] According to embodiments of this disclosure, A′ -1 A′ +1 A and A′0 can be represented as Where R0 can represent the minimum distance between the satellite and the ground target, and n can represent 1 to ur·N r Positive integers.

[0095] According to embodiments of this disclosure, psf(x) can be represented as a point spread function that simulates the energy diffusion of an image along the range direction caused by pulse compression. The point spread function psf(x) can be expressed as follows: (1).

[0096] psf(x)=sinc((2×0.443×x) / (ur×α)) (1)

[0097] Where ur can represent the difference ratio of the amplitude sequence, α can represent the oversampling rate of the distance direction, and x can represent the independent variable, such as time or distance in a certain direction, or the pixel label on a two-dimensional image. For example, if the two-dimensional image is 200*200, then x can represent the independent variable from 0 to 200, where x is an integer.

[0098] According to embodiments of this disclosure, A -1 A +1 A0 and A' can be represented as A' respectively. -1 A′ +1 The final range amplitude distribution sequence obtained after applying the point spread function to A′0 is the final range amplitude sequence, where A′ -1 A′ +1 The value of the independent variable x in the point spread function corresponding to A′0 is different.

[0099] According to embodiments of this disclosure, based on a random amplitude sequence A′ of an inverse Gaussian distribution along the distance direction... -1 and A′ -1 The corresponding point spread function psf(x) yields the final amplitude sequence A in the range direction. -1 It can be expressed as the following formula (2).

[0100] A -1 (R0)=A′ -1 (R0)*psf (2)

[0101] According to embodiments of this disclosure, based on a random amplitude sequence A′ of an inverse Gaussian distribution along the distance direction... +1 and A′ +1 The corresponding point spread function psf(x) yields the final amplitude sequence A in the range direction. +1 It can be expressed as the following formula (3).

[0102] A +1 (R0)=A′ +1 (R0)*psf (3)

[0103] According to an embodiment of this disclosure, the final amplitude sequence A0 in the range direction can be expressed as the following formula (4) based on the inverse Gaussian distribution random amplitude sequence A′0 along the range direction and the point spread function psf(x) corresponding to A′0.

[0104] A0(R0)=A′0(R0)*psf (4)

[0105] In operation S320, a two-dimensional amplitude spectrum is obtained based on the final amplitude sequence in the range direction and the migration line equation corresponding to the final amplitude sequence in the range direction.

[0106] According to embodiments of this disclosure, I′ -1 , I′ +1 I′ and I′0 can be represented as the two-dimensional amplitude spectrum calculated according to the final amplitude sequence in the range direction and the migration line equation corresponding to the final amplitude sequence in the range direction, respectively. -1 , I′ +1 I′0 can also represent the assignment of values ​​to the simulated spectrum according to the migration line equation corresponding to the final amplitude sequence in the range direction.

[0107] According to embodiments of this disclosure, the final amplitude sequence A in the distance direction... -1 The final amplitude sequence A of sum and range direction -1 The two-dimensional amplitude spectrum I′ is obtained by calculating the corresponding fuzzy component migration line equation. -1 It can be expressed as the following formula (5).

[0108] I' -1 (R,f η ) = A -1 (k·f η +Rb), R=1,2,3,…,N r (5)

[0109] Where k and b can represent the final amplitude sequence A in the range direction, respectively. -1 The slope and intercept of the corresponding fuzzy component migration line equation, R can represent the distance between the satellite and the ground target.

[0110] According to embodiments of this disclosure, the final amplitude sequence A in the distance direction... +1 The final amplitude sequence A of sum and range direction +1 The two-dimensional amplitude spectrum I′ is obtained by calculating the corresponding fuzzy component migration line equation. +1 It can be expressed as the following formula (6).

[0111] I' +1 (R,f η ) = A +1 (-k·f η +R+b), R=1,2,3,…,N r (6)

[0112] Where k and b can represent the final amplitude sequence A in the range direction, respectively. +1 The slope and intercept of the corresponding fuzzy component migration line equation, R can represent the distance between the satellite and the ground target.

[0113] According to an embodiment of this disclosure, the two-dimensional amplitude spectrum I′0 calculated according to the final amplitude sequence A0 in the range direction and the principal component migration line equation corresponding to the final amplitude sequence A0 in the range direction can be expressed as the following formula (7).

[0114] I′0(R,f η ) = A0(R), R = 1, 2, 3, ..., N r (7)

[0115] Among them, f η It can represent the azimuth frequency, and R can represent the distance between the satellite and the ground target.

[0116] In operation S330, the two-dimensional amplitude spectrum is subjected to azimuth antenna pattern weighting processing to obtain fuzzy discrimination quantity and main discrimination quantity.

[0117] According to embodiments of this disclosure, W -1 W +1 W0 can be expressed as the azimuth antenna pattern weighting, which can be expressed as the following formula (8).

[0118] W i (f η )=sinc 2 (L / (2v)×(f η -f ηc (8)

[0119] Among them, f η The azimuth frequency can be represented, L can represent the antenna length, v can represent the satellite orbital velocity, and f can represent the satellite orbital velocity. ηc The frequency can represent the Doppler center frequency, where i can represent any value from -1, 0, to +1, and f. η The independent variable in formula (8) can be characterized for the two-dimensional amplitude spectrum I′. -1 , I′ +1 And I′0, in formula (8) f η They are different, for example, for the two-dimensional amplitude spectrum I′ -1 f η The value can range from -400 to 100 Hz; for the two-dimensional amplitude spectrum I′0, f η The value can range from 100 to 500 Hz; for the two-dimensional amplitude spectrum I′ +1 f η The frequency can be between 500 and 900 Hz.

[0120] According to embodiments of this disclosure, the two-dimensional amplitude spectrum I′ -1 Perform azimuth-oriented antenna pattern weighted W -1 Processing yields the fuzzy distinguishing factor I. -1It can be expressed as the following formula (9).

[0121] I -1 (R,f η )=I′ -1 (R,f η )·W -1 (f η (9)

[0122] According to embodiments of this disclosure, the two-dimensional amplitude spectrum I′ +1 Perform azimuth-oriented antenna pattern weighted W +1 Processing yields the fuzzy distinguishing factor I. +1 It can be expressed as the following formula (10).

[0123] I +1 (R,f η )=I′ +1 (R,f η )·W +1 (f η (10)

[0124] According to an embodiment of this disclosure, the two-dimensional amplitude spectrum I′0 is subjected to azimuth antenna pattern weighting W0 processing to obtain the main distinguishing quantity I0, which can be expressed as the following formula (11).

[0125] I0(R,f η )=I′0(R,f η )·W0(f η (11)

[0126] In operation S340, the fuzzy distinguishing factor and the main distinguishing factor are superimposed and coherent noise is added to obtain the simulated spectrum.

[0127] According to embodiments of this disclosure, the fuzzy distinction quantity and the main distinction quantity obtained after weighted processing of the azimuth antenna pattern are superimposed, that is, the fuzzy distinction quantity I... -1 I +1 The signal is simply added to the main signal I0, and then coherent noise is added after the superposition process to obtain the simulated spectrum. The coherent noise can be obtained by interpolation of random Gaussian noise and can be used for coherent superposition of analog signals.

[0128] According to embodiments of this disclosure, the spectral estimation model is trained using a simulated spectrum dataset constructed based on random parameters. Therefore, by introducing random parameters to construct the simulated spectrum dataset, the spectral estimation model can adaptively process SAR images with different imaging parameters even when the parameters are unknown. This allows the spectral estimation model to perform spectral estimation stably and effectively even when parameters are missing or inaccurate.

[0129] According to an embodiment of this disclosure, the method 300 further includes: superimposing a two-dimensional amplitude spectrum to obtain an interference map corresponding to the simulated spectrum; and training a spectrum estimation model based on the simulated spectrum and the interference map corresponding to the simulated spectrum to obtain a trained spectrum estimation model.

[0130] According to an embodiment of this disclosure, the superposition of two-dimensional amplitude spectra can be expressed as the following formula (12).

[0131]

[0132] Where I can represent the interference diagram corresponding to the simulated spectrum.

[0133] According to embodiments of this disclosure, by superimposing two-dimensional amplitude spectra, an interference map corresponding to the simulated spectrum can be obtained. This allows for the training of a spectral estimation model based on the simulated spectrum and the interference map corresponding to the simulated spectrum. The trained spectral estimation model enables the spectral estimation model to adaptively process SAR images with different imaging parameters even when parameters are unknown. This ensures that the spectral estimation model can still perform spectral estimation stably and effectively even when parameters are missing or inaccurate.

[0134] Figure 4 The illustration schematically shows a process flow diagram for constructing a simulated spectrum dataset for training a spectrum estimation model according to an embodiment of the present disclosure.

[0135] like Figure 4 As shown, the simulation spectrum dataset may include simulation spectrum 425 and interference map 426 corresponding to simulation spectrum 425.

[0136] According to an embodiment of this disclosure, using random parameter 401, a random amplitude sequence A′ along the distance direction is randomly generated according to an inverse Gaussian distribution. -1 (i.e., 402), A′0 (i.e., 403), and A′ +1 (i.e., 404). A′ -1 and A′ -1 The corresponding point spread function 405, A′0 and the point spread function 406 corresponding to A′0 and A′ +1 and A′ +1 The corresponding point spread function 407 performs operation S310 respectively, which can obtain the final amplitude sequence A in the range direction. -1 (i.e., 408), A0 (i.e., 409) and A +1 (i.e., 410).

[0137] According to embodiments of this disclosure, A -1 and A -1 The corresponding fuzzy component migration line equations 411, A0, and the principal component migration line equations 412 and A0 are also present.+1 and A +1 The corresponding fuzzy component migration line equation 413, when executed with operation S320, yields the two-dimensional amplitude spectrum I′. -1 (i.e. 414), I′0 (i.e. 415) and I′ +1 (i.e., 416).

[0138] According to embodiments of this disclosure, I′ -1 and I′ -1 The corresponding azimuth weighted antenna direction 417, I′0 and the azimuth weighted antenna direction 418 and I′ corresponding to I′0 +1 and I′ +1 The corresponding azimuth weighting 419 towards the antenna direction, and the execution of operation S330 respectively, can yield the ambiguity resolution I. -1 (i.e., 420), principal distinguisher I0 (i.e., 421), and fuzzy distinguisher I′ +1 (i.e., 422). And for the fuzzy discrimination quantity I... -1 Principal distinguishing factor I0 and fuzzy distinguishing factor I +1 Superposition processing is performed to obtain superimposed component 423, and coherent noise 424 is added to superimposed component 423 to obtain simulated spectrum 425.

[0139] According to embodiments of this disclosure, the two-dimensional amplitude spectrum I′ -1 , I′0 and I′ +1 By superimposing the data using formula (12), the interference diagram 426 corresponding to the simulated spectrum 425 can be calculated.

[0140] Figure 5 A flowchart illustrating the process of obtaining a spectral estimation plot according to an embodiment of the present disclosure is shown schematically.

[0141] like Figure 5 As shown, the method 500 includes operations S510 to S550.

[0142] Figure 6 A schematic diagram of the structure of a spectral estimation model according to an embodiment of the present disclosure is shown.

[0143] like Figure 6As shown, the spectral estimation model can include a multi-scale feature extraction module and a spectral sequence analysis module. The multi-scale feature extraction module can include a first multi-scale feature extraction module and a second multi-scale feature extraction module. The multi-scale feature extraction module can be constructed from a general convolutional neural network, and the spectral sequence analysis module can be constructed from a Long Short-Term Memory (LSTM) module. The multi-scale feature extraction module can perform preliminary parameter evaluation on the range-Doppler domain data and smooth coherent noise, so that the multi-scale feature map initially has a deeper semantic feature. The feature maps at each level and scale are serialized and decomposed, and then input into the spectral sequence analysis module. This module can not only reconstruct the structure of the antenna radiation pattern in the azimuth direction of the main region, but also further analyze the aliased fuzzy energy.

[0144] When operating S510, the distance Doppler domain data is input into the first multi-scale feature extraction module in the spectral estimation model to obtain feature maps of different sizes.

[0145] According to embodiments of this disclosure, the obtained range Doppler domain data is input as follows: Figure 6 In the first multi-scale feature extraction module of the spectral estimation model shown, features are extracted from the range Doppler domain data and then pooled through a pooling layer to obtain feature maps of different sizes. This is because the size of the feature map gradually decreases with each downsampling or pooling operation.

[0146] According to embodiments of this disclosure, such as Figure 6 As shown, feature extraction is performed on the range-Doppler domain data input to the first multi-scale feature extraction module to obtain the features. Figure 1 , for features Figure 1 Downsampling yields features smaller than the feature value. Figure 1 Dimensional characteristics Figure 2 , for features Figure 2 Downsampling yields features smaller than the feature value. Figure 2 Dimensional characteristics Figure 3 …This shows that the feature map size decreases as the number of downsampling iterations increases.

[0147] During operation of S520, feature maps of different sizes are decomposed to obtain the first sequence.

[0148] According to embodiments of this disclosure, feature map decomposition of feature maps of different sizes can be characterized by cutting the obtained feature maps of different sizes into azimuth sequences according to their orientation. A first sequence can represent the azimuth sequence corresponding to a feature map of a corresponding size. For example, feature map decomposition can be equivalent to decomposing a 100*100 feature map into 100 azimuth sequences of length 100, and these 100 azimuth sequences of length 100 are the first sequence corresponding to the 100*100 feature map.

[0149] According to embodiments of this disclosure, such as Figure 6 As shown, for features Figure 1 By performing feature map decomposition, we can obtain features... Figure 1 The corresponding first sequence, for features Figure 2 By performing feature map decomposition, we can obtain features... Figure 2 The corresponding first sequence, for features Figure 3 By performing feature map decomposition, we can obtain features... Figure 3 The corresponding first sequence, and so on, can be used to analyze features. Figure 4 and characteristics Figure 5 Perform feature map decomposition.

[0150] In operation S530, the first sequence is input into the spectral sequence analysis module to obtain the second sequence.

[0151] According to embodiments of this disclosure, the obtained first sequence is input as follows: Figure 6 The corresponding LSTM module in the spectral sequence analysis module, as shown, yields the corresponding second sequence. The LSTM module can be used for sequence analysis, providing estimates of the degree of ambiguity energy interference affecting corresponding sample points in the range-Doppler domain data, as reflected in the corresponding feature map. For example, as... Figure 6 As shown, it will be related to the features Figure 1 The corresponding LSTM module, which takes the first sequence as input, can obtain the features. Figure 1 The corresponding second sequence, from which features can be obtained Figure 1 The second sequence can also be represented as 100 azimuth sequences of length 100, which is an estimate of the degree of interference of the corresponding sample points in the range Doppler domain data.

[0152] In operation S540, the second sequence is reconstructed using a feature map to obtain a reconstructed feature map.

[0153] According to embodiments of this disclosure, the obtained second sequences are spliced ​​and recombined to obtain a recombined feature map. For example, as... Figure 6 As shown, it will be related to the features Figure 1By splicing and recombining the corresponding second sequence, we can obtain the feature... Figure 1 Corresponding recombination features Figure 1 When the second sequence consists of 100 azimuth sequences of length 100, feature map reconstruction of the second sequence yields a reconstructed feature map of size 100*100. The reconstructed feature map has the same size as the original feature map. For example, as shown... Figure 6 As shown, features Figure 1 Recombination characteristics Figure 1 The dimensions are the same.

[0154] When operating the S550, the reconstructed feature map is input into the second multi-scale feature extraction module to obtain the spectral estimation map.

[0155] According to embodiments of this disclosure, inputting the reconstructed feature map into a second multi-scale feature extraction module can restore the reconstructed feature map into a spectral estimation map, for example, as... Figure 6 As shown, the obtained recombination features Figure 1 ~Recombination characteristics Figure 5 The input to the second multi-scale feature extraction module undergoes upsampling, 1×1 convolution, and activation function processing to output a spectral estimation map. The image input to the spectral estimation model and the spectral estimation map output by the model have the same size. For example, with an input image of size 100*100, a spectral estimation map of size 100*100 can be obtained. Each pixel in the spectral estimation map has a corresponding estimated value, reflecting the degree of fuzzy energy interference affecting each sample point in the distance Doppler domain data.

[0156] According to embodiments of this disclosure, by inputting range-Doppler domain data into a spectral estimation model trained based on a simulated spectrum dataset, a spectral estimation map can be obtained that reflects the degree of interference from fuzzy energy at each sample point in the range-Doppler domain data. This enables the spectral estimation model to adaptively process SAR images with different imaging parameters even when parameters are unknown, thus allowing the spectral estimation model to perform spectral estimation stably and effectively even when parameters are missing or inaccurate.

[0157] Figure 7 A flowchart illustrating the extraction of sub-spectral segments according to an embodiment of the present disclosure is shown schematically.

[0158] like Figure 7 As shown, the method 700 includes operations S710 to S730.

[0159] In operation S710, the spectral estimation map is averaged along the range direction to obtain a one-dimensional vector ordered by azimuth frequency.

[0160] According to embodiments of this disclosure, the spectral estimation map E of the obtained range-Doppler domain data is...D Averaging along the range direction yields a one-dimensional vector T ordered by azimuth frequency. A This one-dimensional vector can show the degree of ambiguity energy interference in range-Doppler domain data at different frequency bands. This vector can be used as a basis to select the spectral band with the lowest overall ambiguity energy interference. For example, if the spectral estimation map is 100*100, with the x-axis representing the range direction and the y-axis representing the azimuth direction, averaging the 100 azimuth sequences along the range direction yields 100 one-dimensional vectors ordered by azimuth frequency.

[0161] When operating the S720, the sub-spectral segment is determined based on the one-dimensional vector and the sub-spectral proportion.

[0162] According to embodiments of this disclosure, based on the one-dimensional vector, a corresponding one-dimensional vector T is selected according to the proportion of the subspectrum. A The spectral band least affected by fuzzy energy interference. The subspectral proportion can characterize the ratio of the frequency band containing the subspectral band to the pulse repetition period. This ratio can effectively suppress orientation fuzziness while maintaining the signal-to-noise ratio of the image.

[0163] According to embodiments of this disclosure, the transmission of electromagnetic waves is repeated periodically, with sampling performed each time. This process is discrete, resulting in the received signal undergoing a Fourier transform, limiting its frequency to a finite frequency band. This finite frequency band can also be characterized as a restricted frequency band, and its length is the reciprocal of the pulse repetition period. For example, if the subspectral proportion θ is a fixed value of 22%, then based on a one-dimensional vector, the spectral band least affected by ambiguity energy interference within the corresponding restricted frequency band can be determined. This spectral band accounts for 22% of the corresponding restricted frequency band. Among 100 one-dimensional vectors ordered by azimuth frequency, if the 55th one-dimensional vector shows the least degree of ambiguity energy interference in the range-Doppler domain data at that frequency band, then based on this 55th one-dimensional vector and the subspectral proportion of 22%, the spectral band containing the 34th to 66th one-dimensional vectors is selected as the least affected by ambiguity energy interference, thus determining the subspectral band.

[0164] When operating the S730, the Fourier transform matrix is ​​selected based on the 1A-level complex data and spectrum containing azimuth ambiguity, and sub-spectral segments are extracted.

[0165] According to embodiments of this disclosure, the method of extracting sub-spectral segments can be represented as constructing a corresponding spectral selection Fourier transform matrix F and performing matrix multiplication with 1A-level complex data T.

[0166] According to embodiments of this disclosure, the extracted sub-spectral segments It can represent A complex matrix with r rows and r columns, where, It can be represented as the floor symbol.

[0167] According to an embodiment of this disclosure, the construction of the spectral selection Fourier transform matrix F used to extract sub-spectral segments can be expressed as the following formula (13).

[0168]

[0169] Where, 1∈, 2∈1,2,3···, It can represent a subspectral segment selected based on a one-dimensional vector A and the proportion of the subspectral segment, and it can represent a positive integer number of samples.

[0170] According to embodiments of this disclosure, averaging the spectral estimation map along the distance direction yields a one-dimensional vector ordered by azimuth frequency. Based on the one-dimensional vector and the subspectral proportion, the subspectral segment least affected by azimuth ambiguity energy can be determined. Simultaneously, this subspectral proportion can effectively suppress azimuth ambiguity while maintaining the signal-to-noise ratio of the image. Finally, based on the 1A-level complex data containing azimuth ambiguity and the spectrum, a Fourier transform matrix is ​​selected to extract the subspectral segment.

[0171] According to embodiments of this disclosure, extrapolating a subspectral segment to obtain image data with suppressed azimuth blur includes: obtaining an extrapolation result based on the subspectral segment and the extrapolation formula; and performing an inverse azimuth-to-Fourier transform on the extrapolation result to obtain image data with suppressed azimuth blur.

[0172] According to an embodiment of this disclosure, the extrapolation formula can be expressed as the following formula (14).

[0173]

[0174] Here, can represent an integer index, ranging from 1 to r, which can characterize the i-th column of the matrix along the distance direction; F can represent the constructed spectral selection Fourier transform matrix; E() can represent the i-th column data of the extrapolation result E; and T() can represent the 1A-level complex data Y. T The data in the kth column, It can represent the 1A level complex data Y T The conjugate transpose of the k-th column of data It can represent sub-spectral segments The data in the kth column.

[0175] According to an embodiment of this disclosure, the sub-spectral segment is extrapolated according to the extrapolation formula (14) to obtain the extrapolation result, and the extrapolation result is subjected to azimuth-direction Fourier inverse transform processing to obtain image data with suppressed azimuth blur, so that the image with suppressed azimuth blur has the same azimuth resolution as the original image.

[0176] According to embodiments of this disclosure, image data with suppressed azimuth blur is processed and then input into a target detection model to obtain a final detection result, including: performing amplitude map processing and truncated quantization processing on the image data with suppressed azimuth blur; and inputting the processed image data into a target detection model to obtain a final detection result.

[0177] According to embodiments of this disclosure, the image data for which orientation blur is suppressed is complex data, and the value corresponding to each pixel in the image is a complex number, which may be difficult for general deep learning models to handle. Amplitude mapping can represent taking the modulus of the complex amplitude, thus transforming the complex image data into real image data.

[0178] According to embodiments of this disclosure, the amplitude map itself has a numerical range of 0 to 10,000. However, the remote sensing images used are often large, such as 100*100 images. Among the 10,000 pixels, a few may have higher amplitudes, such as over 9,000 or 10,000, while the others may fall within the 0 to 100 range. Directly processing the amplitude map itself would result in too low contrast. Therefore, the cropping process can represent setting values ​​greater than 100 to 100. For example, if the pixel value of a point is 200, then 200 is set to 100. This cropping process truncates values ​​higher than 100, and the resulting numerical range becomes 1 to 100.

[0179] According to embodiments of this disclosure, quantization processing can characterize a relatively continuous change, for example, quantizing a continuous change such as 0, 0.1, 0.2... to 0~255, which is equivalent to rounding 245.1 to 245, so that the image format is consistent with general image formats such as PNG (Portable Network Graphics), and the image with suppressed orientation blur is treated as an image.

[0180] According to embodiments of this disclosure, the object detection model can be trained using relevant data to obtain a deep learning model, specifically represented as a Faster R-CNN (Faster Region-Convolutional Neural Network) model.

[0181] According to embodiments of this disclosure, the image data with suppressed azimuth blur is subjected to amplitude map processing and truncated quantization processing to facilitate processing by the target detection model. After processing, the image data with suppressed azimuth blur is input into the target detection model, which can solve the problem of false detection and missed detection caused by false targets formed by azimuth blur, thereby obtaining the final detection result.

[0182] Figure 8The schematic diagram illustrates a target detection method for SAR azimuth-ambiguous interference images according to an embodiment of the present disclosure.

[0183] like Figure 8 As shown, the azimuth-to-Fourier transform 802 is performed on the 1A-level complex data 801 containing the azimuth module to obtain the range-Doppler domain data 803, i.e., operation S210 is performed on the 1A-level complex data 801 containing the azimuth module. The range-Doppler domain data 803 is input into the spectral estimation model 804, and operations S510 to S520 are performed to obtain the spectral estimation map 805. Based on the spectral estimation map 805 and the subspectral proportion 806, operations S710 to S730 are performed to obtain the subspectral segment 807. The subspectral segment 807 is extrapolated according to the extrapolation formula 808, i.e., formula (14), to obtain the extrapolation result 809. The extrapolation result 809 is processed by the azimuth-to-Fourier inverse transform 810 to obtain the image data 811 with azimuth blur suppressed. After the image data 811 with suppressed azimuth is processed by amplitude mapping and truncated quantization, it is input into the target detection model 812 to obtain the target detection result 813.

[0184] According to embodiments of this disclosure, the adaptive Wiener filter method is a classic algorithm that suppresses azimuth ambiguity by constructing a selective filter to remove ambiguity energy in the frequency domain of SAR data. It forms the basis of all algorithms for suppressing azimuth ambiguity based on the Wiener filter theoretical framework. This algorithm exhibits excellent performance in suppressing azimuth ambiguity; therefore, it is used as a representative of traditional algorithms and compared with the method proposed in this invention. Figures 9 to 12 below serve as schematic diagrams of the satellite remote sensing data used in the comparative experiments and the comparison results.

[0185] Figure 9 schematically illustrates a diagram of satellite remote sensing data used in a comparative experiment according to an embodiment of the present disclosure.

[0186] As shown in Figure 9, large areas of white “shadows” appear in the images used in the experiment, as shown in Figure 9(a) and Figure 9(b). These “shadows” are directional blurring caused by spectral aliasing.

[0187] Table 1 shows the relevant parameters of the image data used in the comparative experiment, where deg can represent the angle of incidence.

[0188] Table 1

[0189]

[0190]

[0191] According to embodiments of this disclosure, Figure 9(a) can represent image data obtained in strip mode, and Figure 9(b) can represent image data obtained in sliding restraint mode. Using these two images, Figure 9(a) and Figure 9(b), as experimental data, it can be fully demonstrated that the method of the present invention has superior performance compared to conventional methods in all operating modes.

[0192] Figure 10 schematically shows the effect of the orientation blur suppression method of this disclosure in a comparative experiment according to an embodiment of this disclosure.

[0193] As shown in Figure 10, Figure 10(a) can be represented as the effect diagram of suppressing azimuth blur after processing by the azimuth blur suppression method of this disclosure in strip mode; Figure 10(b) can be represented as the effect diagram of suppressing azimuth blur after processing by the azimuth blur suppression method of this disclosure in sliding constraint mode.

[0194] Figure 11 schematically illustrates the effect of processing by a conventional adaptive micro / nano filter method according to an embodiment of the present disclosure.

[0195] As shown in Figure 11, Figure 11(a) can be represented as the effect of suppressing azimuth ambiguity after processing by the traditional adaptive Wiener filter method in strip mode; Figure 11(b) can be represented as the effect of suppressing azimuth ambiguity after processing by the traditional adaptive Wiener filter method in sliding constraint mode.

[0196] According to the embodiments of this disclosure, it is not difficult to find from the comparison of Figures 10(a) and 11(a) and the comparison of Figures 10(b) and 11(b) that the method proposed in this invention has a better effect on suppressing orientation ambiguity than the traditional adaptive Wiener filter method, whether in strip mode or sliding constraint mode.

[0197] Figure 12 schematically illustrates a real test data sample extracted from a test dataset for target detection testing according to an embodiment of the present disclosure.

[0198] According to embodiments of this disclosure, a target detection model, Faster R-CNN, is trained and tested using a SAR image dataset. The training and test sets are independent and do not contain each other. A set of test results is obtained by training and testing the target detection model without any processing of the dataset. Then, the orientation blur suppression method of this invention is applied to the training and test sets for image optimization, and the target detection model is retrained and tested to obtain another set of test results. A comparison of the two sets of target detection performance results is shown in Table 2.

[0199] Table 2

[0200] Detection methods Recall rate accuracy General methods 89.17％ 88.99％ The method proposed in this invention 96.20％ 95.74％

[0201] According to the embodiments of this disclosure, the results comparison shown in Table 2 demonstrates that the intelligent optimization and target detection integrated method proposed in this invention significantly improves both recall and accuracy. Figures 12(a) and 12(b) respectively illustrate the visualization of the detection results of the general detection method and the detection method of this invention on the same image.

[0202] According to the embodiments of this disclosure, there are a large number of "shadows" in Figure 12(a), so there are instances of missed or false detections of targets. Correctly detected targets are marked with solid rectangles, falsely detected targets are marked with dashed rectangles, and missed targets are circled. Figure 12(b) uses the integrated image optimization and target detection method of this invention, so the "shadows" are suppressed, the image quality is optimized and improved, and the target detection results are more accurate.

[0203] Based on the above-described target detection method for SAR azimuth-blurred interference images, this disclosure also provides a target detection device for SAR azimuth-blurred interference images. The following will be combined with... Figure 13 The device is described in detail.

[0204] Figure 13 A schematic block diagram of a target detection apparatus for SAR azimuth-ambiguous interference images according to an embodiment of the present disclosure is shown.

[0205] like Figure 13 As shown, the target detection device 1300 for SAR azimuth-ambiguous interference images in this embodiment includes a first acquisition module 1310, a second acquisition module 1320, a capture module 1330, a third acquisition module 1340, and a fourth acquisition module 1350.

[0206] The first acquisition module 1310 is used to perform azimuth-to-Fourier transform processing on the 1A-level complex data containing azimuth blur to obtain range-Doppler domain data, wherein the 1A-level complex data containing azimuth blur represents the original image data affected by azimuth blur interference. In one embodiment, the first acquisition module 1310 can be used to perform the operation S210 described above, which will not be repeated here.

[0207] The second acquisition module 1320 is used to input range-Doppler domain data into the spectral estimation model to obtain a spectral estimation map. The spectral estimation model is trained based on a simulated spectrum dataset, which is constructed using random parameters. The simulated spectrum dataset includes a simulated spectrum and an interference map corresponding to the simulated spectrum. The spectral estimation model includes a multi-scale feature extraction module and a spectral sequence analysis module. In one embodiment, the second acquisition module 1320 can be used to perform the operation S220 described above, which will not be repeated here.

[0208] The extraction module 1330 is used to extract sub-spectral segments based on the spectral estimation diagram and the sub-spectral proportion, wherein the sub-spectral proportion represents the ratio of the frequency band where the sub-spectral segment is located to the pulse repetition period. In one embodiment, the extraction module 1330 can be used to perform the operation S230 described above, which will not be repeated here.

[0209] The third acquisition module 1340 is used to extrapolate the subspectral bands to obtain image data with suppressed azimuth blur, wherein the image with suppressed azimuth blur has the same azimuth resolution as the original image. In one embodiment, the third acquisition module 1340 can be used to perform the operation S240 described above, which will not be repeated here.

[0210] The fourth acquisition module 1350 is used to process the image data with suppressed orientation blur and input it into the target detection model to obtain the final detection result. In one embodiment, the fourth acquisition module 1350 can be used to perform the operation S250 described above, which will not be repeated here.

[0211] According to embodiments of this disclosure, the second obtaining module 1320 includes a first obtaining unit, a second obtaining unit, a first processing unit, and a second processing unit.

[0212] The first obtaining unit is used to obtain the final amplitude sequence in the range direction based on the inverse Gaussian distribution random amplitude sequence along the range direction and the point spread function, wherein the inverse Gaussian distribution random amplitude sequence along the range direction is generated based on random parameters.

[0213] The second obtaining unit is used to obtain the two-dimensional amplitude spectrum based on the final amplitude sequence in the range direction and the migration line equation corresponding to the final amplitude sequence in the range direction.

[0214] The first processing unit is used to perform azimuth antenna pattern weighting processing on the two-dimensional amplitude spectrum to obtain fuzzy discrimination and main discrimination.

[0215] The second processing unit is used to superimpose the fuzzy distinguishing factor and the main distinguishing factor and add coherent noise to obtain the simulated spectrum.

[0216] According to embodiments of this disclosure, the second acquisition module 1320 further includes a third processing unit and a training unit.

[0217] The third processing unit is used to superimpose the two-dimensional amplitude spectrum to obtain an interference map corresponding to the simulated spectrum.

[0218] The training unit is used to train the spectrum estimation model based on the simulated spectrum and the interference map corresponding to the simulated spectrum, and obtain the trained spectrum estimation model.

[0219] According to embodiments of this disclosure, the second obtaining module 1320 further includes a first input unit, a disassembly unit, a second input unit, a reassembly unit, and a third input unit.

[0220] The first input unit is used to input the distance Doppler domain data into the first multi-scale feature extraction module in the spectral estimation model to obtain feature maps of different sizes.

[0221] The decomposition unit is used to decompose feature maps of different sizes to obtain the first sequence.

[0222] The second input unit is used to input the first sequence into the spectral sequence analysis module to obtain the second sequence.

[0223] The recombination unit is used to recombine the feature map of the second sequence to obtain a recombined feature map.

[0224] The third input unit is used to input the reconstructed feature map into the second multi-scale feature extraction module to obtain the spectral estimation map.

[0225] According to embodiments of this disclosure, the interception module 1330 includes a third obtaining unit, a determining unit, and an interception unit.

[0226] The third acquisition unit is used to average the spectrum estimation map along the range direction to obtain a one-dimensional vector ordered by azimuth frequency.

[0227] The unit of determination is used to determine the subspectral segment based on the one-dimensional vector and the subspectral proportion.

[0228] The truncation unit is used to select the Fourier transform matrix based on the 1A-level complex data and spectrum containing azimuth ambiguity, and to truncate sub-spectral segments.

[0229] According to embodiments of this disclosure, the third obtaining module 1340 includes a fourth obtaining unit and a fourth processing unit.

[0230] The fourth obtaining unit is used to obtain the extrapolation result based on the sub-spectral segment and the extrapolation formula.

[0231] The fourth processing unit is used to perform an inverse Fourier transform of the extrapolation results to obtain image data with suppressed azimuth blur.

[0232] According to embodiments of this disclosure, the fourth obtaining module 1350 includes a fifth processing unit and a fourth input unit.

[0233] The fifth processing unit is used to perform amplitude mapping and truncated quantization on the image data with suppressed azimuth blur.

[0234] The fourth input unit is used to input the processed image data into the target detection model to obtain the final detection result.

[0235] According to embodiments of this disclosure, any plurality of modules among the first obtaining module 1310, the second obtaining module 1320, the interception module 1330, the third obtaining module 1340, and the fourth obtaining module 1350 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the first obtaining module 1310, the second obtaining module 1320, the interception module 1330, the third obtaining module 1340, and the fourth obtaining module 1350 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in a suitable combination of any of these. Alternatively, at least one of the first obtaining module 1310, the second obtaining module 1320, the intercepting module 1330, the third obtaining module 1340, and the fourth obtaining module 1350 may be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0236] Figure 14 A block diagram of an electronic device suitable for implementing a target detection method for SAR azimuth-blurred images is illustrated according to an embodiment of the present disclosure.

[0237] like Figure 14 As shown, an electronic device 1400 according to an embodiment of the present disclosure includes a processor 1401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1402 or a program loaded from a storage portion 1408 into a random access memory (RAM) 1403. The processor 1401 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1401 may also include onboard memory for caching purposes. The processor 1401 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0238] RAM 1403 stores various programs and data required for the operation of electronic device 1400. Processor 1401, ROM 1402, and RAM 1403 are interconnected via bus 1404. Processor 1401 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1402 and / or RAM 1403. It should be noted that the programs may also be stored in one or more memories other than ROM 1402 and RAM 1403. Processor 1401 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0239] According to embodiments of this disclosure, the electronic device 1400 may further include an input / output (I / O) interface 1405, which is also connected to a bus 1404. The electronic device 1400 may also include one or more of the following components connected to the I / O interface 1405: an input section 1406 including a keyboard, mouse, etc.; an output section 1407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1408 including a hard disk, etc.; and a communication section 1409 including a network interface card such as a LAN card, modem, etc. The communication section 1409 performs communication processing via a network such as the Internet. A drive 1410 is also connected to the I / O interface 1405 as needed. A removable medium 1411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1410 as needed so that computer programs read from it can be installed into the storage section 1408 as needed.

[0240] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0241] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 1402 and / or RAM 1403 and / or one or more memories other than ROM 1402 and RAM 1403 described above.

[0242] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the item recommendation method provided in the embodiments of this disclosure.

[0243] When the computer program is executed by the processor 1401, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0244] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1409, and / or installed from the removable medium 1411. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0245] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1409, and / or installed from the removable medium 1411. When the computer program is executed by the processor 1401, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0246] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0247] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0248] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0249] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A target detection method for SAR azimuth ambiguity interfered images, comprising: performing azimuth Fourier transform processing on 1A-level complex data containing azimuth ambiguity to obtain range-Doppler domain data, wherein the 1A-level complex data containing azimuth ambiguity represents original image data interfered by azimuth ambiguity; inputting the range-Doppler domain data into a spectrum estimation model to obtain a spectrum estimation graph, wherein the spectrum estimation model is obtained by training according to a simulation spectrum data set, the simulation spectrum data set is constructed by using random parameters, the simulation spectrum data set includes a simulation spectrum and an interference graph corresponding to the simulation spectrum, and the spectrum estimation model includes a multi-scale feature extraction module and a spectrum sequence analysis module; wherein the simulation spectrum data set is constructed by using random parameters, including: obtaining a final range amplitude sequence according to an inverse Gaussian distribution random amplitude sequence along the range direction and a point spread function, wherein the inverse Gaussian distribution random amplitude sequence along the range direction is generated according to the random parameters; obtaining a two-dimensional amplitude spectrum according to the final range amplitude sequence and a shift line equation corresponding to the final range amplitude sequence; performing azimuth antenna pattern weighting processing on the two-dimensional amplitude spectrum to obtain a blur component and a main component; and performing superposition processing and adding coherent noise processing on the blur component and the main component to obtain the simulation spectrum; according to the spectrum estimation graph and a sub-spectrum proportion, intercepting a sub-spectrum segment, wherein the sub-spectrum proportion represents a ratio of a frequency band where the sub-spectrum segment is located to a pulse repetition period; performing extrapolation on the sub-spectrum segment to obtain image data in which the azimuth ambiguity is suppressed, wherein the image data in which the azimuth ambiguity is suppressed has the same azimuth resolution as the original image; inputting the image data in which the azimuth ambiguity is suppressed after processing into a target detection model to obtain a final detection result.

2. The method of claim 1, further comprising: performing superposition processing on the two-dimensional amplitude spectrum to obtain an interference graph corresponding to the simulation spectrum; and training the spectrum estimation model according to the simulation spectrum and the interference graph corresponding to the simulation spectrum to obtain the trained spectrum estimation model. the multi-scale feature extraction module includes a first multi-scale feature extraction module and a second multi-scale feature extraction module, and the inputting the range-Doppler domain data into the spectrum estimation model to obtain a spectrum estimation graph includes:

3. The method of claim 1, wherein, inputting the range-Doppler domain data into the first multi-scale feature extraction module in the spectrum estimation model to obtain feature maps of different sizes; performing feature map disassembly on the feature maps of different sizes to obtain a first sequence; inputting the first sequence into the spectrum sequence analysis module to obtain a second sequence; performing feature map recombination on the second sequence to obtain a recombined feature map; and inputting the recombined feature map into the second multi-scale feature extraction module to obtain the spectrum estimation graph. the according to the spectrum estimation graph and a sub-spectrum proportion, intercepting a sub-spectrum segment includes:

4. The method of claim 1, wherein, averaging the spectrum estimation graph along the range direction to obtain a one-dimensional vector in order of azimuth frequency; ​ determine the sub-spectrum segment according to the one-dimensional vector and the sub-spectrum proportion; and According to the 1A order complex data containing azimuth ambiguity and the spectrum selection Fourier transform matrix, the sub-spectrum segment is intercepted.

5. The method of claim 1, wherein, The extrapolation of the sub-spectrum segment obtains the image data in which the azimuth ambiguity is suppressed, comprising: According to the sub-spectrum segment and the extrapolation formula, an extrapolation result is obtained; and The azimuth Fourier inverse transform processing is performed on the extrapolation result to obtain the image data in which the azimuth ambiguity is suppressed.

6. The method of claim 1, wherein, The image data in which the azimuth ambiguity is suppressed is processed and input into a target detection model to obtain a final detection result, comprising: The image data in which the azimuth ambiguity is suppressed is processed to obtain the final detection result.

7. A target detection device for SAR azimuth ambiguity interference image, comprising: A first obtaining module is used for performing azimuth Fourier transform processing on 1A order complex data containing azimuth ambiguity to obtain range Doppler domain data, wherein the 1A order complex data containing azimuth ambiguity represents original image data interfered by azimuth ambiguity; A second obtaining module is used for inputting the range Doppler domain data into a spectrum estimation model to obtain a spectrum estimation graph, wherein the spectrum estimation model is obtained by training according to a simulation spectrum data set, the simulation spectrum data set is constructed by using random parameters, the simulation spectrum data set includes a simulation spectrum and an interference graph corresponding to the simulation spectrum, and the spectrum estimation model includes a multi-scale feature extraction module and a spectrum sequence analysis module; The second obtaining module includes a first obtaining unit, a second obtaining unit, a first processing unit and a second processing unit; The first obtaining unit is used for obtaining a final amplitude sequence along a range direction according to an inverse Gaussian distribution random amplitude sequence along the range direction and a point spread function, wherein the inverse Gaussian distribution random amplitude sequence along the range direction is generated according to the random parameters; The second obtaining unit is used for obtaining a two-dimensional amplitude spectrum according to the final amplitude sequence along the range direction and a shift line equation corresponding to the final amplitude sequence along the range direction; The first processing unit is used for performing azimuth antenna pattern weighting processing on the two-dimensional amplitude spectrum to obtain a blur component and a main component; The second processing unit is used for performing superposition processing and adding coherent noise processing on the blur component and the main component to obtain the simulation spectrum; An intercepting module is used for intercepting a sub-spectrum segment according to the spectrum estimation graph and a sub-spectrum proportion, wherein the sub-spectrum proportion represents a ratio of a frequency band in which the sub-spectrum segment is located to a pulse repetition period; A third obtaining module is used for extrapolating the sub-spectrum segment to obtain image data in which the azimuth ambiguity is suppressed, wherein the image data in which the azimuth ambiguity is suppressed has the same azimuth resolution as the original image.

8. An electronic device, comprising: one or more processors; a storage device for storing one or more programs, ​ ​ wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to carry out the method according to any one of claims 1-6.

9. A computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1-6.

Images