Moving target detection method and device, equipment and medium

Abstract

The invention provides a moving target detection method which can be applied to the technical field of radar signal processing and artificial intelligence. The method comprises the following steps: carrying out sub-aperture segmentation on full-aperture spaceborne synthetic aperture radar echo data in an azimuth direction to obtain multiple pieces of sub-aperture data; performing distance compression, phase compensation and backward projection imaging processing on each piece of sub-aperture data to generate a sub-aperture image sequence; processing the sub-aperture image sequence by using a pre-trained deep learning semantic segmentation model to obtain a pixel-level segmentation mask of the moving target; and inverting the peak position of the moving target in the echo signal after distance compression based on the pixel-level segmentation mask. The invention also provides a moving target detection device, equipment and a medium.

Description

Technical Field

[0001] This invention relates to the fields of radar signal processing and artificial intelligence technology, and more specifically to a method, apparatus, device and medium for detecting moving targets. Background Technology

[0002] With the continuous development of spaceborne synthetic aperture radar (SAR) technology, the new generation of satellite systems has acquired the capability for long-duration continuous observation, providing an important technical means for the continuous monitoring of moving targets. However, under long-duration synthetic aperture observation conditions, the detection of moving targets faces severe challenges. Due to the complex relative motion between the target and the radar platform, moving targets often exhibit severe defocusing and positional shifts in SAR images, forming complex trajectories spanning multiple pixels. This significantly limits the performance of traditional detection methods.

[0003] Currently, moving target detection mainly employs two technical approaches: echo domain and image domain. Echo domain detection methods (such as constant false alarm rate detection) rely on accurate modeling of the statistical characteristics of background clutter. However, under long-term observation conditions, sea states are complex and variable, and target echo energy significantly attenuates, making it difficult to establish accurate statistical models. Furthermore, these methods typically only utilize one-dimensional coherent accumulation in the range direction, failing to effectively distinguish weak targets from background clutter. Image domain detection methods face different challenges. Detection methods based on wake features fail for targets moving from land towards the sea surface, while detection methods based on multi-channel systems significantly increase system complexity and cost. Simultaneously, image processing methods based on background subtraction are extremely sensitive to image registration accuracy; registration errors directly lead to an increased false alarm rate.

[0004] More fundamentally, existing echo-domain and image-domain detection methods lack an effective correlation mechanism. While echo-domain detection preserves complete signal physical information, direct detection is difficult; image-domain detection provides an intuitive geometric representation but loses phase details, making signal inversion challenging. This technical barrier limits further improvements in moving target detection accuracy and hinders subsequent signal-level processing and analysis. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address at least one of the aforementioned problems, this invention provides a moving target detection method, apparatus, device, and medium. By introducing a processing framework combining sub-aperture segmentation and back projection imaging, and integrating deep learning semantic segmentation technology, it achieves pixel-level accurate detection and cross-domain information inversion of moving targets under long-term observation conditions of spaceborne SAR. The method first divides the full aperture data into multiple sub-apertures for independent imaging, significantly enhances the energy of moving targets through two-dimensional coherent accumulation, and then uses a semantic segmentation model to extract pixel-level spatial features of moving targets from the sub-aperture image sequence to achieve accurate positioning and segmentation. Finally, based on the segmentation results, the peak position of the moving target signal in the echo domain is inverted. Thus, while maintaining the low complexity of the single-channel system, it effectively solves the technical problems of severe target defocusing, insufficient detection accuracy, and information disconnect between the image domain and the echo domain under long synthetic aperture conditions.

[0007] (II) Technical Solution

[0008] To address the aforementioned technical problems, embodiments of the present invention provide a method, apparatus, device, and medium for detecting moving targets.

[0009] According to a first aspect of the present invention, a moving target detection method is provided, comprising: performing sub-aperture segmentation on full-aperture spaceborne synthetic aperture radar echo data in the azimuth direction to obtain multiple sub-aperture data; performing range compression, phase compensation, and back projection imaging processing on each sub-aperture data to generate a sub-aperture image sequence; processing the sub-aperture image sequence using a pre-trained deep learning semantic segmentation model to obtain a pixel-level segmentation mask of the moving target; and retrieving the peak position of the moving target in the range-compressed echo signal based on the pixel-level segmentation mask.

[0010] In some exemplary embodiments, sub-aperture segmentation of full-aperture spaceborne synthetic aperture radar echo data in the azimuth direction includes: determining the number and length of sub-apertures based on the synthetic aperture time and the motion characteristics of the moving target; and dividing the full-aperture data uniformly or non-uniformly into multiple sub-apertures in the azimuth direction based on the number and length of the sub-apertures, with each sub-aperture containing a predetermined number of azimuth pulse samples.

[0011] In some exemplary embodiments, performing range compression, phase compensation, and back projection imaging processing on each sub-aperture data includes: establishing a non-flat slant-viewing imaging geometric model for spaceborne synthetic aperture radar; calculating the instantaneous slant range between the satellite and the imaging grid points at each azimuth time; constructing a phase compensation function based on the instantaneous slant range; performing phase compensation on the range-compressed signal; and coherently accumulating the compensated signals at all azimuth times to generate a sub-aperture image sequence.

[0012] In some exemplary embodiments, the training method for the deep learning semantic segmentation model includes: acquiring a sample set of sub-aperture images of a spaceborne synthetic aperture radar containing moving targets, the sample set containing pixel-level annotations; constructing a semantic segmentation network with an encoder-decoder structure; training the semantic segmentation network using the sample set, optimizing the parameters of the semantic segmentation network by minimizing the loss function; saving the trained network parameters to obtain a pre-trained deep learning semantic segmentation model; wherein the loss function includes a cross-entropy loss function; and the moving targets in the sample set include targets with defocus features and trajectory offset features.

[0013] In some exemplary embodiments, processing a sub-aperture image sequence using a pre-trained deep learning semantic segmentation model to obtain a pixel-level segmentation mask for a moving target includes: inputting the sub-aperture image sequence into the pre-trained deep learning semantic segmentation model; extracting multi-scale features of the image through the encoder part, including shallow and deep features, where the shallow features contain detail information and the deep features contain semantic information; upsampling and feature fusion of the multi-scale features through the decoder part to obtain the spatial resolution of the feature map; processing the output features using an activation function in the last layer of the decoder to generate a probability map of each pixel belonging to a moving target; and obtaining the final pixel-level segmentation mask based on the probability map through thresholding, where pixels with a probability value greater than a set threshold are determined to be moving targets.

[0014] In some exemplary embodiments, retrieving the peak position of a moving target in the range-compressed echo signal based on a pixel-level segmentation mask includes: determining the pixel position of the moving target in the sub-aperture image according to the segmentation mask; calculating the instantaneous slant range of the imaging grid point corresponding to the pixel position relative to the satellite at each azimuth time; calculating the corresponding time delay value based on the instantaneous slant range; and determining the peak position of the moving target echo signal after range compression based on the time delay value.

[0015] In some exemplary embodiments, determining the peak position of the moving target echo signal after range compression based on the time delay value includes: calculating the time delay difference between the imaging grid point where the moving target is located and the nearest distance reference point; converting the time delay difference into a range gate index according to the system's fast time sampling interval; and determining the peak position of the moving target in the range compressed signal based on the range gate index.

[0016] According to a second aspect of the present invention, a moving target detection device is provided, comprising: a data segmentation module for performing sub-aperture segmentation on full-aperture spaceborne synthetic aperture radar echo data in the azimuth direction to obtain multiple sub-aperture data; an image generation module for performing range compression, phase compensation, and back projection imaging processing on each sub-aperture data to generate a sub-aperture image sequence; a mask acquisition module for processing the sub-aperture image sequence using a pre-trained deep learning semantic segmentation model to obtain a pixel-level segmentation mask of the moving target; and a target detection module for retrieving the peak position of the moving target in the range-compressed echo signal based on the pixel-level segmentation mask.

[0017] According to a third aspect of the present invention, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0018] A fourth aspect of the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0019] (III) Beneficial Effects

[0020] As can be seen from the above technical solutions, the moving target detection method, apparatus, device, and medium provided by the embodiments of the present invention have at least the following beneficial effects:

[0021] (1) By using sub-aperture segmentation and back projection imaging, two-dimensional coherent accumulation of echo signals is achieved, which effectively enhances the energy of moving targets in the image domain and significantly improves the signal-to-clutter ratio of targets and clutter, thereby enabling reliable detection of moving targets in complex backgrounds.

[0022] (2) By introducing a deep learning semantic segmentation model, the spatial distribution features of moving targets in sub-aperture image sequences are fully utilized to achieve pixel-level accurate segmentation of moving targets, accurately identify the complete trajectory of moving targets, and overcome the positioning deviation problem caused by target defocus in traditional methods.

[0023] (3) Based on the pixel-level detection results in the image domain, the peak position of the moving target signal in the echo domain is accurately retrieved by calculating the slant distance and time delay information of the imaging grid points corresponding to the moving target. This is the first time that the inversion from the image domain detection to the echo domain signal has been realized, providing complete data support for subsequent signal-level processing such as motion parameter estimation.

[0024] (4) The present invention adopts a single-channel system combined with a sub-aperture processing strategy, which eliminates the need for a complex hardware architecture that requires multi-channel collaborative operation. It avoids additional image registration operations through unified imaging grid processing, and significantly reduces system complexity and implementation cost while maintaining high performance.

[0025] (5) The deep learning semantic segmentation framework adopted does not depend on a specific network structure and can be adapted to different model architectures according to actual application needs. It is also applicable to moving target detection in various scenarios such as sea surface and land-sea interface, showing good versatility and adaptability. Attached Figure Description

[0026] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0027] Figure 1 A flowchart illustrating a moving target detection method according to an embodiment of the present invention is shown schematically;

[0028] Figure 2 This diagram illustrates the architecture of a pixel-level moving target detection method with a sub-aperture number of 3.

[0029] Figure 3 A moving target detection device according to an embodiment of the present invention is illustrated schematically; and

[0030] Figure 4 A block diagram of an electronic device for a moving target detection method according to an embodiment of the present invention is shown schematically. Detailed Implementation

[0031] Hereinafter, embodiments of the present invention will 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 invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention 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 concept of the invention.

[0032] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0033] 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.

[0034] 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 those 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.).

[0035] Figure 1 A flowchart illustrating a moving target detection method according to an embodiment of the present invention is shown schematically; Figure 2 The diagram illustrates the architecture of a pixel-level moving target detection method with a sub-aperture number of 3.

[0036] like Figure 1 and Figure 2 As shown, a moving target detection method according to an embodiment of the present invention includes steps S110 to S140.

[0037] In step S110, the full-aperture spaceborne synthetic aperture radar echo data is divided into sub-apertures in the azimuth direction to obtain multiple sub-aperture data.

[0038] In some exemplary embodiments, step S110 may specifically include: determining the number and length of sub-apertures based on the synthetic aperture time and the motion characteristics of the moving target; and dividing the full aperture data into multiple sub-apertures uniformly or non-uniformly in the azimuth direction based on the number and length of the sub-apertures, with each sub-aperture containing a predetermined number of azimuth pulse samples.

[0039] Step S110 effectively reduces the synthesis aperture time within each sub-aperture by dividing the long synthetic aperture into multiple short-time sub-apertures. This process significantly reduces the defocusing degree of the moving target in each sub-aperture image, making the target energy more concentrated; it effectively shortens the target's motion trajectory within a single sub-aperture, preserving the target's motion change characteristics in the azimuth direction; and it provides a temporally related image sequence for subsequent processing, laying the foundation for detection using target motion features.

[0040] For example, in actual data processing, satellites in Position in a coordinate system is obtained through Obtained through navigation. An imaging grid is established on the Earth's surface based on latitude and longitude parameters, and subsequent imaging processing is performed within this grid.

[0041] Distance time axis set The orientation time axis is set to In terms of location and time At that time, satellite The coordinates are represented as The coordinates of the imaging grid points are represented as follows: In terms of location and time At that time, the instantaneous distance between the satellite and the imaging grid points It can be represented as follows:

[0042] (1)

[0043] Establishing a long synthetic aperture time-based spaceborne system in a geodetic coordinate system Non-planar oblique-view imaging geometry. Setting the origin of the geodetic coordinate system. The center of the imaging grid. Always moving target Located at the origin of the geodetic coordinate system , The axis points due east to the horizon. The axis points due north. The axis satisfies the right-hand screw law, and the target's velocity vector... During the imaging time (tens of seconds), the satellite's motion is approximately uniform linear motion, with the velocity vector being... Satellite position vector . Distance history with satellites It can be represented as:

[0044] (2)

[0045] Long Synthetic Aperture Time Spaceborne Aperture segmentation is performed on the echo data. The full aperture is divided into... Each sub-aperture contains [number] sub-apertures. Each direction pulse sample.

[0046] In step S120, distance compression, phase compensation, and back projection imaging are performed on each sub-aperture data to generate a sub-aperture image sequence.

[0047] In some exemplary embodiments, step S120 may specifically include: establishing a non-flat slant-view imaging geometric model for spaceborne synthetic aperture radar; calculating the instantaneous slant range between the satellite and the imaging grid points at each azimuth time; constructing a phase compensation function based on the instantaneous slant range; performing phase compensation on the range-compressed signal; and coherently accumulating the compensated signals at all azimuth times to generate a sub-aperture image sequence.

[0048] Step S120 converts the echo data into image domain information through precise imaging processing. Its beneficial effects are as follows: Two-dimensional coherent accumulation of the echo signal is achieved through range compression and back projection algorithms, significantly improving the signal-to-noise ratio; phase compensation technology effectively corrects phase errors caused by platform motion and target movement, ensuring imaging quality; the generated high-quality sub-aperture image sequence preserves the target's scattering characteristics and reveals the target's temporal variation patterns, providing ideal input data for deep learning detection.

[0049] For example, after segmenting the echo data, based on the long synthetic aperture time-spaceborne... Non-flat oblique-view imaging geometry is used to demodulate and compress the echo signal within each sub-aperture. After demodulation and range compression, the echo signal within each sub-aperture... echo signal It can be represented as follows:

[0050] (3)

[0051] in, It is the azimuth time within an azimuth sub-aperture. For radar wavelength, At the speed of light, It is an azimuth window function. For bandwidth.

[0052] Based on long synthetic aperture time spaceborne Non-flat oblique imaging geometry model, positional shift caused by target motion for:

[0053] (4)

[0054] in Satellite position vector Projected to Vectors in a plane. velocity vector Projected to Vectors in a plane.

[0055] Based on long synthetic aperture time spaceborne Non-flat oblique imaging geometry model, quadratic phase error caused by target motion for:

[0056] (5)

[0057] in The time for synthesizing the aperture is denoted as .

[0058] At each azimuth time within the sub-aperture, the slant range between the radar and each imaging grid point can be calculated according to equation (2), denoted as . And from this, the corresponding time delay value is obtained. .

[0059] At each azimuth time within the sub-aperture, the time delay of each imaging grid point is calculated sequentially, and phase compensation is performed on the signal. Compensation function. for:

[0060] (6)

[0061] Phase compensation of the signal is performed sequentially at each azimuth time within the sub-aperture, and the results within each sub-aperture are summed to obtain the imaging result of that sub-aperture. :

[0062] (7)

[0063] In step S130, the sub-aperture image sequence is processed using a pre-trained deep learning semantic segmentation model to obtain a pixel-level segmentation mask for the moving target.

[0064] In some exemplary embodiments, the training method for the deep learning semantic segmentation model may include: acquiring a sample set of sub-aperture images of a spaceborne synthetic aperture radar containing moving targets, the sample set containing pixel-level annotations; constructing a semantic segmentation network with an encoder-decoder structure; training the semantic segmentation network using the sample set, optimizing the parameters of the semantic segmentation network by minimizing a loss function; saving the trained network parameters to obtain a pre-trained deep learning semantic segmentation model; wherein the loss function includes a cross-entropy loss function; and the moving targets in the sample set include targets with defocus features and trajectory offset features.

[0065] In some exemplary embodiments, step S130 may specifically include: inputting the sub-aperture image sequence into a pre-trained deep learning semantic segmentation model; extracting multi-scale features of the image through the encoder part, including shallow and deep features, where the shallow features contain detail information and the deep features contain semantic information; upsampling and feature fusion of the multi-scale features through the decoder part to obtain the spatial resolution of the feature map; processing the output features using an activation function in the last layer of the decoder to generate a probability map of each pixel belonging to a moving target; and obtaining the final pixel-level segmentation mask based on the probability map through thresholding, where pixels with a probability value greater than a set threshold are determined to be moving targets.

[0066] Step S130 achieves accurate target detection through deep learning technology. Its beneficial effects include: leveraging the powerful feature learning capabilities of deep learning models to automatically extract discriminative features such as defocus and displacement of moving targets in sub-aperture image sequences; the semantic segmentation model achieves pixel-level accurate detection, accurately identifying the contours and positions of moving targets; and the end-to-end processing method avoids the complex manual feature design and parameter adjustment required in traditional methods, improving the method's adaptability and robustness.

[0067] For example, with Taking the network as an example, based on The network is trained to obtain a model suitable for segmenting moving targets, and then... The segmentation model can obtain a segmentation map of the moving target.

[0068] go through After processing by the encoder and decoder, a pixel-level segmentation mask for the sub-aperture moving target can be obtained. For segmentation, use The function calculates the logarithmic loss score, which is mainly used to update network weights during training. The output is the sub-aperture segmentation result. It can be represented as:

[0069] (8)

[0070] in, and These are the convolutional kernel weights and biases of the output layer. It is an activation function used to calculate probabilities.

[0071] In step S140, the peak position of the moving target in the range-compressed echo signal is retrieved based on the pixel-level segmentation mask.

[0072] In some exemplary embodiments, step S140 may specifically include: determining the pixel position of the moving target in the sub-aperture image based on the segmentation mask; calculating the instantaneous slant range of the imaging grid point corresponding to the pixel position relative to the satellite at each azimuth time; calculating the corresponding time delay value based on the instantaneous slant range; and determining the peak position of the moving target echo signal after range compression based on the time delay value.

[0073] In some exemplary embodiments, determining the peak position of the moving target echo signal after range compression based on the time delay value includes: calculating the time delay difference between the imaging grid point where the moving target is located and the nearest distance reference point; converting the time delay difference into a range gate index according to the system's fast time sampling interval; and determining the peak position of the moving target in the range compressed signal based on the range gate index.

[0074] Step S140 realizes the inversion from the image domain detection result to the echo domain signal. Its beneficial effects are mainly reflected in: establishing the correspondence between the image domain detection and the echo domain signal, providing accurate guiding information for subsequent signal-level processing; through peak position inversion, the time-frequency characteristics of the moving target in the echo data can be accurately determined, providing a data basis for motion parameter estimation; and realizing the organic combination of detection and measurement, forming a complete processing chain, and improving the practical value of the entire system.

[0075] For example, based on the results of sub-aperture pixel-level image segmentation, the pixel position of the moving target in the image can be obtained. Calculate the distance between the imaging grid point containing the moving target pixel and the satellite at each azimuth time within the sub-aperture. :

[0076] (9)

[0077] Calculate the time delay of the grid point containing the moving target pixel at each time point in each orientation. :

[0078] (10)

[0079] Calculate the time delay of the imaging grid point containing the moving target pixel relative to the nearest reference point. :

[0080] (11)

[0081] in The slope distance is the closest reference point. The duration of the pulse.

[0082] Calculate the peak position of the echo signal distance after compression for a moving target at different azimuth times. :

[0083] (12)

[0084] in For fast sampling intervals.

[0085] For each sub-aperture echo data, perform sub-aperture analysis separately. Imaging processing is performed to obtain a pixel-level segmentation map of the moving target at the sub-aperture. Based on the pixel-level segmentation results of the moving target at the sub-aperture, the peak position corresponding to the distance between the echo data of each sub-aperture and the compressed moving target is calculated.

[0086] Figure 3 The illustration schematically depicts a moving target detection device according to an embodiment of the present invention.

[0087] like Figure 3 As shown, the moving target detection device 800 of this embodiment includes a data segmentation module 810, an image generation module 820, a mask acquisition module 830, and a target detection module 840.

[0088] The data segmentation module 810 is used to perform sub-aperture segmentation on the azimuth direction of the full-aperture spaceborne synthetic aperture radar echo data to obtain multiple sub-aperture data.

[0089] The image generation module 820 is used to perform distance compression, phase compensation and back projection imaging processing on each sub-aperture data to generate a sub-aperture image sequence.

[0090] The mask acquisition module 830 is used to process the sub-aperture image sequence using a pre-trained deep learning semantic segmentation model to obtain the pixel-level segmentation mask of the moving target.

[0091] The target detection module 840 is used to retrieve the peak position of a moving target in the range-compressed echo signal based on a pixel-level segmentation mask.

[0092] According to embodiments of the present invention, any plurality of modules among the data segmentation module 810, image generation module 820, mask acquisition module 830, and target detection module 840 may be combined into one module, or any one of these modules may be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules may be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the data segmentation module 810, image generation module 820, mask acquisition module 830, and target detection module 840 may 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 any other reasonable means of integrating or packaging circuitry, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the data segmentation module 810, image generation module 820, mask acquisition module 830, and target detection module 840 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0093] Figure 4 A block diagram of an electronic device for a moving target detection method according to an embodiment of the present invention is shown schematically.

[0094] like Figure 4As shown, an electronic device 900 according to an embodiment of the present invention includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 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 901 may also include onboard memory for caching purposes. The processor 901 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 invention.

[0095] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 902 and / or RAM 903. It should be noted that programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in one or more memories.

[0096] According to an embodiment of the present invention, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0097] The present invention 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, which, when executed, implement the method according to the embodiments of the present invention.

[0098] According to embodiments of the present invention, a 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 the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, a computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0099] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

Claims

1. A moving object detection method characterized by, The method comprises: sub-aperture segmentation of full-aperture spaceborne synthetic aperture radar echo data in the azimuth direction to obtain multiple sub-aperture data; distance compression, phase compensation and back-projection imaging processing are respectively performed on each of the sub-aperture data to generate a sub-aperture image sequence; a pre-trained deep learning semantic segmentation model is used to process the sub-aperture image sequence to obtain a pixel-level segmentation mask of the moving target; and based on the pixel-level segmentation mask, the peak position of the moving target in the distance-compressed echo signal is inverted.

2. The method of claim 1, wherein, The sub-aperture segmentation of full-aperture spaceborne synthetic aperture radar echo data in the azimuth direction comprises: determining the number and length of sub-apertures according to the synthetic aperture time and the motion characteristics of the moving target; based on the number and length of the sub-apertures, the full-aperture data is uniformly or non-uniformly divided into multiple sub-apertures in the azimuth direction, and each sub-aperture contains a predetermined number of azimuth pulse samples.

3. The method of claim 1, wherein, The distance compression, phase compensation and back-projection imaging processing of each of the sub-aperture data comprises: establishing a spaceborne synthetic aperture radar non-flat squint imaging geometric model; calculating the instantaneous slant range between the satellite and the imaging grid point at each azimuth time; constructing a phase compensation function based on the instantaneous slant range; phase compensation is performed on the distance-compressed signal, and the compensated signals at all azimuth times are coherently accumulated to generate a sub-aperture image sequence.

4. The method of claim 1, wherein, The training method of the deep learning semantic segmentation model comprises: obtaining a spaceborne synthetic aperture radar sub-aperture image sample set containing a moving target, the sample set containing pixel-level labeling; constructing a semantic segmentation network with an encoder-decoder structure; training the semantic segmentation network using the sample set to optimize the parameters of the semantic segmentation network by minimizing the loss function; saving the trained network parameters to obtain the pre-trained deep learning semantic segmentation model; wherein the loss function comprises a cross-entropy loss function; the moving target in the sample set includes targets with defocusing characteristics and trajectory offset characteristics.

5. The method of claim 4, wherein, The use of a pre-trained deep learning semantic segmentation model to process the sub-aperture image sequence to obtain a pixel-level segmentation mask of the moving target comprises: inputting the sub-aperture image sequence into the pre-trained deep learning semantic segmentation model; extracting multi-scale features of the image through the encoder part, the multi-scale features including shallow features and deep features, wherein the shallow features contain detailed information and the deep features contain semantic information; performing upsampling and feature fusion on the multi-scale features through the decoder part to obtain the spatial resolution of the feature map; using an activation function to process the output features at the last layer of the decoder to generate a probability map of each pixel belonging to the moving target; based on the probability map, a threshold processing is performed to obtain the final pixel-level segmentation mask, wherein the pixels with a probability value greater than a set threshold are determined as the moving target.

6. The method of claim 1, wherein, The inversion of the peak position of the moving target in the distance-compressed echo signal based on the pixel-level segmentation mask comprises: determining the pixel position of the moving target in the sub-aperture image according to the segmentation mask; calculating an instantaneous slant range of the imaging grid point corresponding to the pixel position relative to the satellite at each azimuth time instant; calculating a corresponding time delay value according to the instantaneous slant range; determining a peak position of the echo signal of the moving target after range compression based on the time delay value.

7. The method of claim 6, wherein, The determining of the peak position of the echo signal of the moving target after range compression based on the time delay value comprises: calculating a time delay difference of the imaging grid point where the moving target is located relative to the nearest range reference point; converting the time delay difference into a range gate index according to a system fast time sampling interval; determining a peak position of the moving target in the range compressed signal based on the range gate index.

8. A moving object detection apparatus characterized by comprising: The device comprises: a data segmentation module configured to perform sub-aperture segmentation on full-aperture spaceborne synthetic aperture radar echo data in the azimuth direction to obtain a plurality of sub-aperture data; an image generation module configured to perform range compression, phase compensation and back-projection imaging processing on each of the sub-aperture data respectively to generate a sub-aperture image sequence; a mask acquisition module configured to process the sub-aperture image sequence using a pre-trained deep learning semantic segmentation model to obtain a pixel-level segmentation mask of the moving target; and a target detection module configured to inversely derive a peak position of the echo signal of the moving target after range compression based on the pixel-level segmentation mask.

9. An electronic device, comprising: comprise: one or more processors; a memory for storing one or more computer programs, the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, the computer program or instructions, when executed by a processor, implement the steps of the method according to any one of claims 1-7.

Images