Distance Doppler domain reconstruction method, device, equipment and medium

Abstract

The invention provides a distance Doppler domain reconstruction method, device and equipment and a medium, and belongs to the technical field of radar, and the method comprises the steps: carrying out the pulse emission based on a predetermined pulse emission moment, and collecting a non-uniform echo signal; inputting the non-uniform echo signal into a pre-trained distance Doppler domain (RD) reconstruction network to obtain an RD reconstruction result; the training process of the RD reconstruction network comprises the steps of generating a non-uniform echo signal sample according to a pulse emission time sample, obtaining a reconstruction image according to the non-uniform echo signal sample and a preset reconstruction network, and constructing a loss function according to the pulse emission time sample and the reconstruction image, and adjusting parameters of the pulse emission time sample and the reconstruction network according to the loss function. According to the method, the pulse emission time and the RD reconstruction technology are combined, so that the parameters of the RD reconstruction technology can be fully combined with the characteristics of a non-uniform pulse emission mode, the maximum unambiguous detection distance expansion is realized, and the RD reconstruction quality is improved.

Description

Technical Field

[0001] The present invention relates to the field of radar technology, and in particular to a range Doppler domain reconstruction method, device, equipment and medium. Background Art

[0002] Pulse Doppler (PD) radar is a type of radar that uses the Doppler effect to measure the radial motion speed of an object in the direction of the radar beam. Range Doppler (RD) reconstruction is an important step in extracting target distance and speed, which can provide support for meteorological observation, track tracking, battlefield reconnaissance, etc.

[0003] In PD radar, there is a set of inherent constraints between the maximum unambiguous detection range and the maximum unambiguous Doppler. Many scholars have achieved the extension of the maximum unambiguous detection range by increasing the minimum time interval between adjacent pulses. However, the existing technology has not optimized the RD reconstruction algorithm in a targeted manner and cannot fully resist aliasing, which has reduced the RD reconstruction performance and accuracy to a certain extent. Summary of the invention

[0004] The present invention provides a range Doppler domain reconstruction method, device, equipment and medium, which solve the defect that the prior art cannot fully resist aliasing and reduces the RD reconstruction performance and accuracy.

[0005] The present invention provides a range Doppler domain reconstruction method, comprising: Perform pulse transmission based on a predetermined pulse transmission time to collect a non-uniform echo signal; Inputting the non-uniform echo signal into a pre-trained range Doppler domain RD reconstruction network to obtain an RD reconstruction result; Among them, the training process of the RD reconstruction network includes generating non-uniform echo signal samples according to pulse emission time samples, obtaining a reconstructed image according to the non-uniform echo signal samples and a preset reconstruction network, constructing a loss function according to the pulse emission time samples and the reconstructed image, adjusting the respective parameters of the pulse emission time samples and the reconstruction network according to the loss function, and repeating the training steps until a preset convergence condition is reached.

[0006] As an embodiment, obtaining a reconstructed image according to the non-uniform echo signal sample and a preset reconstruction network includes: Performing range-direction matched filtering on the non-uniform echo signal samples; Based on the reconstruction network, linear reconstruction and nonlinear filtering are performed for a preset number of iterations on the filtered non-uniform echo signal samples to obtain the reconstructed image.

[0007] As an embodiment, the step of performing linear reconstruction and nonlinear filtering for a preset number of iterations on the filtered non-uniform echo signal samples based on the reconstruction network to obtain the reconstructed image includes: Use the zero matrix as the initialization to reconstruct the image; Determine a non-uniform inverse Fourier transform matrix and a non-uniform Fourier transform matrix according to the filtered non-uniform echo signal samples; Inputting the initialized reconstructed image, the non-uniform inverse Fourier transform matrix and the non-uniform Fourier transform matrix into the reconstruction network to obtain the reconstructed image output by the reconstruction network; The reconstruction network is used to determine the linear reconstruction result of the current iteration in each iteration process according to the non-uniform inverse Fourier transform matrix, the non-uniform Fourier transform matrix, the filtered non-uniform echo signal samples and the initialized reconstructed image / reconstruction result of the previous iteration, perform non-linear filtering on the linear reconstruction result of the current iteration to obtain the reconstruction result of the current iteration, repeat a preset number of iterations to obtain the reconstructed image.

[0008] As an embodiment, performing nonlinear filtering on the linear reconstruction result of the current iteration to obtain the reconstruction result of the current iteration includes: Based on the soft threshold iterative function, the linear reconstruction result of the current iteration is nonlinearly filtered to obtain the reconstruction result of the current iteration.

[0009] As an embodiment, constructing a loss function according to the pulse emission time samples and the reconstructed image includes: The loss function is constructed according to the gap between the reconstructed image and the real image and the minimum value constraint of the samples at the pulse emission moment.

[0010] As an embodiment, the expression of the loss function is as follows: ; in, is the loss function, For the n The reconstructed image corresponding to the sample at the time of pulse emission, N is the total number of samples, For the n The real image corresponding to the sample at the time of pulse emission, is the first sample vector at the time of pulse emission m numerical values, M is the total number of sample vector values ​​at the time of pulse emission.

[0011] As an embodiment, adjusting the pulse emission time sample and the parameters of the reconstruction network according to the loss function includes: Generate a set of mask sequences corresponding to the pulse emission time samples based on a random parameter freezing strategy; Multiplying the gradient of the loss function with respect to the sample at the pulse emission time by the mask sequence to achieve random zeroing; According to the gradient of the loss function after random zeroing to the sample at the pulse emission time, the random part parameters of the sample at the pulse emission time and the parameters of the reconstruction network are updated.

[0012] The present invention also provides a range Doppler domain reconstruction device, comprising: A training module, used to generate a non-uniform echo signal sample according to a pulse emission time sample, obtain a reconstructed image according to the non-uniform echo signal sample and a preset reconstruction network, construct a loss function according to the pulse emission time sample and the reconstructed image, adjust the parameters of the pulse emission time sample and the reconstruction network according to the loss function, and repeat the training steps until a preset convergence condition is reached; An acquisition module, used for performing pulse emission based on a predetermined pulse emission time to acquire a non-uniform echo signal; The reconstruction module is used to input the non-uniform echo signal into a pre-trained range Doppler domain RD reconstruction network to obtain an RD reconstruction result.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the above-mentioned range Doppler domain reconstruction methods is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the range Doppler domain reconstruction method described above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the range Doppler domain reconstruction method described above is implemented.

[0016] The range Doppler domain reconstruction method, device, equipment and medium provided by the present invention perform pulse emission based on a predetermined pulse emission time to collect non-uniform echo signals; input the non-uniform echo signals into a pre-trained range Doppler domain RD reconstruction network to obtain RD reconstruction results; wherein the training process of the RD reconstruction network includes generating non-uniform echo signal samples according to pulse emission time samples, obtaining reconstructed images according to the non-uniform echo signal samples and a preset reconstruction network, constructing a loss function according to the pulse emission time samples and the reconstructed images, adjusting the parameters of the pulse emission time samples and the reconstruction network according to the loss function, and repeating the training steps until the preset convergence conditions are reached. The present invention combines the pulse emission time with the RD reconstruction technology, so that the parameters of the RD reconstruction technology can fully combine the characteristics of a specific non-uniform pulse emission mode, and optimize the RD reconstruction algorithm in a targeted manner, which can fully resist aliasing and improve the reconstruction quality of RD while achieving the maximum unambiguous detection distance extension. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is one of the flow charts of the range Doppler domain reconstruction method provided by the present invention.

[0019] Figure 2 This is the second flow chart of the range Doppler domain reconstruction method provided by the present invention.

[0020] Figure 3 It is a schematic diagram of the training process of the RD reconstruction network provided by the present invention.

[0021] Figure 4a It is a schematic diagram of the reconstruction effect of the existing range Doppler domain reconstruction method. Figure 4b It is a schematic diagram of the reconstruction effect of the range Doppler domain reconstruction method provided by the present invention.

[0022] Figure 5 It is a structural schematic diagram of the range Doppler domain reconstruction device provided by the present invention.

[0023] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] It should be noted that all actions of acquiring signals, information or data in the present invention are performed in compliance with the corresponding data protection laws and policies of the location and with the authorization given by the owner of the corresponding device.

[0026] The non-uniform pulse emission mode can avoid the spectrum aliasing caused by the decrease of pulse repetition frequency to a certain extent, but there is no targeted optimization of the reconstruction algorithm in the RD domain, resulting in the reconstruction algorithm being unable to fully utilize the anti-aliasing capability of the non-uniform emission mode, which reduces the reconstruction performance and accuracy of the RD domain to a certain extent.

[0027] In order to overcome the problem of inability to fully remove spectral aliasing due to the complete decoupling of the existing PD radar RD reconstruction algorithm and the non-uniform emission pattern design, the present invention proposes a joint optimization strategy, which combines the learnable pulse emission pattern and the RD reconstruction algorithm for joint data-driven joint optimization to obtain high-quality RD reconstruction results, which can improve the quality of RD reconstruction while achieving the maximum unambiguous detection distance extension.

[0028] Figure 1 is one of the flow charts of the range Doppler domain reconstruction method provided by the present invention, Figure 2 FIG. 2 is a flow chart of the range Doppler domain reconstruction method provided by the present invention. Figure 1 and Figure 2 As shown, the present invention provides a range Doppler domain reconstruction method, including steps S100 to S200.

[0029] Step S100, pulse emission is performed based on a predetermined pulse emission time, and a non-uniform echo signal is collected. The pulse emission time is the optimal pulse emission time determined during the RD reconstruction network training process. The pulse emission sequence can be determined based on the pulse emission time. The pulse emission sequence refers to a group of instantaneous pulse signals that are regularly emitted in time and in a certain order. The non-uniform echo signal refers to the non-uniform pulse echo data obtained by the PD radar after pulse emission and echo collection according to the pulse emission time.

[0030] Step S200: input the non-uniform echo signal into a pre-trained range Doppler domain RD reconstruction network to obtain an RD reconstruction result.

[0031] Among them, the training process of the RD reconstruction network includes generating non-uniform echo signal samples according to pulse emission time samples, obtaining a reconstructed image according to the non-uniform echo signal samples and a preset reconstruction network, constructing a loss function according to the pulse emission time samples and the reconstructed image, adjusting the respective parameters of the pulse emission time samples and the reconstruction network according to the loss function, and repeating the training steps until a preset convergence condition is reached.

[0032] The repeated training step refers to returning to and executing the step of generating non-uniform echo signal samples according to the pulse emission time samples. After reaching the preset convergence condition, the pulse emission time samples are used as the optimal pulse emission time, and the reconstruction network is used as the RD reconstruction network.

[0033] The method provided by the present invention includes a training process and a use process. The training process refers to the training process of the RD reconstruction network to determine the optimal pulse emission sequence and the matching RD reconstruction algorithm. The training refers to step S100 and step S200. The trained pulse emission sequence is used to transmit pulses and collect echoes. After the non-uniform pulse echo data is collected, the trained RD reconstruction network is used to obtain the final reconstruction result.

[0034] It can be understood that the present invention combines the pulse emission time and the RD reconstruction technology, so that the parameters of the RD reconstruction technology can fully combine the characteristics of the specific non-uniform pulse emission mode, and perform targeted optimization of the RD reconstruction algorithm, which can fully resist aliasing and improve the reconstruction quality of RD while achieving the maximum unambiguous detection distance extension.

[0035] Based on the above embodiment, as an optional embodiment, the generating of non-uniform echo signal samples according to pulse emission time samples includes the following steps.

[0036] Step S011, substitute the preset system parameters of the PD radar, the preset simulation scenario and the pulse emission time samples into the echo simulation model to obtain the simulated non-uniform echo signal samples.

[0037] The pulse emission time sample includes fixed parameters and random parameters. The embodiment of the present invention uses the random parameter part of the pulse emission time sample as a learnable variable.

[0038] It can be understood that the present invention uses the random parameter part in the pulse emission time as a learnable variable, which can increase the randomness and robustness of the non-uniform sequence and improve the anti-aliasing performance.

[0039] Based on the above embodiment, as an optional embodiment, obtaining a reconstructed image according to the non-uniform echo signal samples and a preset reconstruction network includes the following steps.

[0040] Step S021, performing range-oriented matched filtering on the non-uniform echo signal samples. Range-oriented matched filtering is fast time-domain matched filtering.

[0041] Step S022: performing linear reconstruction and nonlinear filtering for a preset number of iterations on the filtered non-uniform echo signal samples based on the reconstruction network to obtain the reconstructed image.

[0042] Optionally, performing linear reconstruction and nonlinear filtering for a preset number of iterations on the filtered non-uniform echo signal samples based on the reconstruction network to obtain the reconstructed image includes the following steps.

[0043] Step S0221, using the zero matrix as an initialization to reconstruct the image.

[0044] Step S0222: Determine a non-uniform inverse Fourier transform matrix and a non-uniform Fourier transform matrix according to the filtered non-uniform echo signal samples. The non-uniform inverse Fourier transform matrix is ​​a slow time domain non-uniform inverse Fourier transform matrix, and the non-uniform Fourier transform matrix is ​​a slow time domain non-uniform Fourier transform matrix.

[0045] Step S0223, inputting the initialized reconstructed image, the non-uniform inverse Fourier transform matrix and the non-uniform Fourier transform matrix into the reconstruction network to obtain the reconstructed image output by the reconstruction network.

[0046] The reconstruction network is used to determine the linear reconstruction result of the current iteration in each iteration process according to the non-uniform inverse Fourier transform matrix, the non-uniform Fourier transform matrix, the filtered non-uniform echo signal samples and the initialized reconstructed image / reconstruction result of the previous iteration, perform non-linear filtering on the linear reconstruction result of the current iteration to obtain the reconstruction result of the current iteration, repeat a preset number of iterations to obtain the reconstructed image.

[0047] Optionally, performing nonlinear filtering on the linear reconstruction result of the current iteration to obtain the reconstruction result of the current iteration includes: performing nonlinear filtering on the linear reconstruction result of the current iteration based on a soft threshold iterative function to obtain the reconstruction result of the current iteration.

[0048] It can be understood that the present invention provides an expanded RD reconstruction algorithm based on the depth of approximate observation, which can greatly reduce the time complexity of the algorithm, and solves the problem that the existing RD reconstruction algorithm for achieving maximum unambiguous detection distance extension only designs a non-uniform pulse emission mode, and simply connects the RD reconstruction algorithm based on compressed sensing in series to obtain the reconstruction result, and does not consider the problem of targeted design caused by the spectrum aliasing caused by a specific emission mode.

[0049] Based on the above embodiment, as an optional embodiment, the construction of the loss function according to the pulse emission time samples and the reconstructed image includes the following steps.

[0050] Step S031, constructing the loss function according to the gap between the reconstructed image and the real image and the minimum value constraint of the samples at the pulse emission moment.

[0051] Specifically, the expression of the loss function is as follows: ; in, is the loss function, For the n The reconstructed image corresponding to the sample at the time of pulse emission, N is the total number of samples, For the n The real image corresponding to the sample at the time of pulse emission, is the mth value of the sample vector at the pulse emission time, and M is the total number of values ​​of the sample vector at the pulse emission time.

[0052] It can be understood that the present invention constructs a loss function through the pulse emission time samples and the reconstructed image, and determines a joint optimization target, so as to simultaneously optimize the pulse emission time and the learnable parameters in the RD reconstruction algorithm, obtain the optimal pulse emission time and the matching RD reconstruction algorithm, and improve the quality of RD reconstruction while achieving the maximum unambiguous detection distance extension.

[0053] Based on the above embodiment, as an optional embodiment, adjusting the pulse emission time sample and the respective parameters of the reconstruction network according to the loss function includes the following steps.

[0054] Step S041: Generate a set of mask sequences corresponding to the pulse emission time samples based on the random parameter freezing strategy. Each element of the mask sequence is a random variable that obeys the Bernoulli distribution.

[0055] Step S042: multiplying the gradient of the loss function with respect to the sample at the pulse emission moment by the mask sequence to achieve random zeroing.

[0056] Step S043, updating the random part parameters of the pulse emission time samples and the parameters of the reconstruction network according to the gradient of the loss function after random zeroing for the pulse emission time samples.

[0057] It can be understood that the present invention introduces a strategy of random parameter freezing during back propagation, by randomly setting a part of the elements in the updated gradient of the parameters at the pulse emission moment to zero, so as to increase the randomness and robustness of the non-uniform sequence and improve the anti-aliasing performance.

[0058] Figure 3 Schematic diagram of the training process of the RD reconstruction network provided by the present invention, such as Figure 3 As shown, the training input parameters of the RD reconstruction network include the PD simulation scene X, which is , PD system parameters And the number of iterations (The present invention is set to 11).

[0059] The training process of the RD reconstruction network specifically includes the following steps: Step 1: According to PD system parameters and simulation scene X, generating a simulated non-uniform echo signal Y.

[0060] Step 2: Perform range matching filtering on the echo signal Y: ; ; in, is the intermediate result after distance matching filtering, is the distance matched filter function, is the distance sampling frequency, is the range frequency modulation, FFT is the Fourier transform, and IFFT is the inverse Fourier transform.

[0061] Step 3: Initialize RD reconstruction results is a zero matrix.

[0062] Step 4: The reconstruction network includes 11 reconstruction network units, each of which includes a linear module and a nonlinear module.

[0063] Linear module: Using the slow-time domain non-uniform inverse Fourier transform matrix Get the matrix after approximate distance matching filtering , calculate the approximate echo matrix and the true distance to the matched filter The difference between them is then used to obtain the approximate reconstruction result of the residual according to the slow-time domain non-uniform Fourier transform matrix, which is added to the imaging result of the previous round. On the top, we get the reconstruction result of the linear module, that is, .

[0064] Step 5: Nonlinear module: Use the soft threshold iterative function to filter out noise, that is , is the soft threshold iterative function.

[0065] Step 6: Repeat steps 4 and 5 for a total of times, and obtain the final reconstructed image .

[0066] Step 7: Calculate the loss function: The loss function is mainly divided into two parts. The former is the main optimization target of the joint optimization, that is, the gap between the final RD reconstruction result and the label Label, and the latter is the minimum value constraint for the pulse emission interval. The expression is as follows: ; in, is the loss function, For the n The reconstructed image corresponding to the sample at the time of pulse emission, N is the total number of samples, For the n The real image corresponding to the sample at the time of pulse emission, is the mth value of the sample vector at the pulse emission time, and M is the total number of values ​​of the sample vector at the pulse emission time.

[0067] Step 8: Use the random parameter freezing strategy to randomly generate a set of mask sequences with the same length as the pulse emission time , where each element is a random variable that follows the Bernoulli distribution. The mask sequence and the Loss function are used to calculate the pulse emission time. Multiply the gradient of , in order to achieve the purpose of random zeroing.

[0068] Step 9: Update the parameters according to the gradient, and repeat steps 1-9 until convergence to obtain the final optimized pulse emission time. and matching RD reconstruction algorithm.

[0069] Figure 4a It is a schematic diagram of the reconstruction effect of the existing range Doppler domain reconstruction method. Figure 4b : is a schematic diagram of the reconstruction effect of the range Doppler domain reconstruction method provided by the present invention. The following table is a comparison table of reconstruction result indicators of the present invention and the prior art. According to the accompanying drawings and tables, it can be seen that the present invention greatly improves the reconstruction accuracy of the range Doppler domain.

[0070] In summary, the present invention proposes a data-driven joint optimization method, which combines a learnable pulse emission mode with an RD reconstruction algorithm, and takes the final RD reconstruction result as the optimization target. Through the joint optimization of the two, the parameters of the reconstruction algorithm can fully combine the characteristics of a specific non-uniform pulse emission mode, and improve the reconstruction quality of RD while achieving the maximum unambiguous detection distance extension. On the one hand, the random part in the pulse emission mode is used as a learnable variable, and a random parameter freezing strategy is introduced during back propagation. By randomly setting a part of the elements in the updated gradient of the pulse emission sequence parameters to zero, the randomness and robustness of the non-uniform sequence are increased, and the anti-aliasing performance is improved. On the other hand, the time complexity of the algorithm can be greatly reduced, and it is expected that the PD radar speed measurement can achieve a longer detection range and higher accuracy.

[0071] The range Doppler domain reconstruction device provided by the present invention is described below. The range Doppler domain reconstruction device described below and the range Doppler domain reconstruction method described above can be referred to each other.

[0072] Figure 5 Schematic diagram of the structure of the range Doppler domain reconstruction device provided by the present invention. Figure 5 As shown, the present invention also provides a range Doppler domain reconstruction device, including the following modules.

[0073] A training module 510 is used to generate a non-uniform echo signal sample according to a pulse emission time sample, obtain a reconstructed image according to the non-uniform echo signal sample and a preset reconstruction network, construct a loss function according to the pulse emission time sample and the reconstructed image, adjust the parameters of the pulse emission time sample and the reconstruction network according to the loss function, and repeat the training steps until a preset convergence condition is reached; The acquisition module 520 is used to perform pulse emission based on a predetermined pulse emission time to acquire a non-uniform echo signal; The reconstruction module 530 is used to input the non-uniform echo signal into a pre-trained range Doppler domain RD reconstruction network to obtain an RD reconstruction result.

[0074] As an embodiment, the training module 510 is further used for: Performing range-direction matched filtering on the non-uniform echo signal samples; Based on the reconstruction network, linear reconstruction and nonlinear filtering are performed for a preset number of iterations on the filtered non-uniform echo signal samples to obtain the reconstructed image.

[0075] As an embodiment, the training module 510 is further used for: Use the zero matrix as the initialization to reconstruct the image; Determine a non-uniform inverse Fourier transform matrix and a non-uniform Fourier transform matrix according to the filtered non-uniform echo signal samples; Inputting the initialized reconstructed image, the non-uniform inverse Fourier transform matrix and the non-uniform Fourier transform matrix into the reconstruction network to obtain the reconstructed image output by the reconstruction network; The reconstruction network is used to determine the linear reconstruction result of the current iteration in each iteration process according to the non-uniform inverse Fourier transform matrix, the non-uniform Fourier transform matrix, the filtered non-uniform echo signal samples and the initialized reconstructed image / reconstruction result of the previous iteration, perform non-linear filtering on the linear reconstruction result of the current iteration to obtain the reconstruction result of the current iteration, repeat a preset number of iterations to obtain the reconstructed image.

[0076] As an embodiment, the training module 510 is further used for: Based on the soft threshold iterative function, the linear reconstruction result of the current iteration is nonlinearly filtered to obtain the reconstruction result of the current iteration.

[0077] As an embodiment, the training module 510 is further used for: The loss function is constructed according to the gap between the reconstructed image and the real image and the minimum value constraint of the samples at the pulse emission moment.

[0078] As an embodiment, the expression of the loss function is as follows: ; in, is the loss function, For the n The reconstructed image corresponding to the sample at the time of pulse emission, N is the total number of samples, For the n The real image corresponding to the sample at the time of pulse emission, is the mth value of the sample vector at the pulse emission time, and M is the total number of values ​​of the sample vector at the pulse emission time.

[0079] As an embodiment, the training module 510 is further used for: Generate a set of mask sequences corresponding to the pulse emission time samples based on a random parameter freezing strategy; Multiplying the gradient of the loss function with respect to the sample at the pulse emission time by the mask sequence to achieve random zeroing; According to the gradient of the loss function after random zeroing to the sample at the pulse emission time, the random part parameters of the sample at the pulse emission time and the parameters of the reconstruction network are updated.

[0080] It should be noted that the range Doppler domain reconstruction device provided by the present invention can execute the range Doppler domain reconstruction method described in any of the above embodiments during specific operation, and has the technical effect corresponding to the method, which will not be described in detail in this embodiment.

[0081] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the range Doppler domain reconstruction method, which includes: performing pulse emission based on a predetermined pulse emission time to collect non-uniform echo signals; inputting the non-uniform echo signals into a pre-trained range Doppler domain RD reconstruction network to obtain RD reconstruction results; wherein the training process of the RD reconstruction network includes generating non-uniform echo signal samples according to pulse emission time samples, obtaining a reconstructed image according to the non-uniform echo signal samples and a preset reconstruction network, constructing a loss function according to the pulse emission time samples and the reconstructed image, adjusting the parameters of the pulse emission time samples and the reconstruction network according to the loss function, and repeating the training steps until a preset convergence condition is reached.

[0082] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0083] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the range Doppler domain reconstruction method provided by the above methods, and the method includes: performing pulse emission based on a predetermined pulse emission time to collect non-uniform echo signals; inputting the non-uniform echo signals into a pre-trained range Doppler domain RD reconstruction network to obtain RD reconstruction results; wherein the training process of the RD reconstruction network includes generating non-uniform echo signal samples based on pulse emission time samples, obtaining reconstructed images based on the non-uniform echo signal samples and a preset reconstruction network, constructing a loss function based on the pulse emission time samples and the reconstructed image, adjusting the respective parameters of the pulse emission time samples and the reconstruction network according to the loss function, and repeating the training steps until a preset convergence condition is reached.

[0084] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the range Doppler domain reconstruction method provided by the above-mentioned methods, the method comprising: performing pulse emission based on a predetermined pulse emission time to collect a non-uniform echo signal; inputting the non-uniform echo signal into a pre-trained range Doppler domain RD reconstruction network to obtain an RD reconstruction result; wherein the training process of the RD reconstruction network comprises generating a non-uniform echo signal sample based on a pulse emission time sample, obtaining a reconstructed image based on the non-uniform echo signal sample and a preset reconstruction network, constructing a loss function based on the pulse emission time sample and the reconstructed image, adjusting the respective parameters of the pulse emission time sample and the reconstruction network according to the loss function, and repeating the training steps until a preset convergence condition is reached.

[0085] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0086] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A range Doppler domain reconstruction method, characterized in that: include: Perform pulse transmission based on a predetermined pulse transmission time to collect a non-uniform echo signal; Inputting the non-uniform echo signal into a pre-trained range Doppler domain RD reconstruction network to obtain an RD reconstruction result; Among them, the training process of the RD reconstruction network includes generating non-uniform echo signal samples according to pulse emission time samples, obtaining a reconstructed image according to the non-uniform echo signal samples and a preset reconstruction network, constructing a loss function according to the pulse emission time samples and the reconstructed image, adjusting the respective parameters of the pulse emission time samples and the reconstruction network according to the loss function, and repeating the training steps until a preset convergence condition is reached.

2. The range Doppler domain reconstruction method according to claim 1, characterized in that: The step of obtaining a reconstructed image according to the non-uniform echo signal sample and a preset reconstruction network includes: Performing range-direction matched filtering on the non-uniform echo signal samples; Based on the reconstruction network, linear reconstruction and nonlinear filtering are performed for a preset number of iterations on the filtered non-uniform echo signal samples to obtain the reconstructed image.

3. The range Doppler domain reconstruction method according to claim 2, characterized in that: The step of performing linear reconstruction and nonlinear filtering for a preset number of iterations on the filtered non-uniform echo signal samples based on the reconstruction network to obtain the reconstructed image includes: Use the zero matrix as the initialization to reconstruct the image; Determine a non-uniform inverse Fourier transform matrix and a non-uniform Fourier transform matrix according to the filtered non-uniform echo signal samples; Inputting the initialized reconstructed image, the non-uniform inverse Fourier transform matrix and the non-uniform Fourier transform matrix into the reconstruction network to obtain the reconstructed image output by the reconstruction network; The reconstruction network is used to determine the linear reconstruction result of the current iteration in each iteration process according to the non-uniform inverse Fourier transform matrix, the non-uniform Fourier transform matrix, the filtered non-uniform echo signal samples and the initialized reconstructed image / reconstruction result of the previous iteration, perform non-linear filtering on the linear reconstruction result of the current iteration to obtain the reconstruction result of the current iteration, repeat a preset number of iterations to obtain the reconstructed image.

4. The range Doppler domain reconstruction method according to claim 3, characterized in that: The performing nonlinear filtering on the linear reconstruction result of the current iteration to obtain the reconstruction result of the current iteration includes: Based on the soft threshold iterative function, the linear reconstruction result of the current iteration is nonlinearly filtered to obtain the reconstruction result of the current iteration.

5. The range Doppler domain reconstruction method according to claim 1, characterized in that: The constructing a loss function according to the pulse emission time samples and the reconstructed image comprises: The loss function is constructed according to the gap between the reconstructed image and the real image and the minimum value constraint of the samples at the pulse emission moment.

6. The range Doppler domain reconstruction method according to claim 5, characterized in that: The expression of the loss function is as follows: ； in, is the loss function, For the n The reconstructed image corresponding to the sample at the time of pulse emission, N is the total number of samples, For the n The real image corresponding to the sample at the time of pulse emission, is the mth value of the sample vector at the pulse emission time, and M is the total number of values ​​of the sample vector at the pulse emission time.

7. The range Doppler domain reconstruction method according to claim 1 or 5, characterized in that: The adjusting the parameters of the pulse emission time sample and the reconstruction network according to the loss function comprises: Generate a set of mask sequences corresponding to the pulse emission time samples based on a random parameter freezing strategy; Multiplying the gradient of the loss function with respect to the sample at the pulse emission time by the mask sequence to achieve random zeroing; According to the gradient of the loss function after random zeroing to the sample at the pulse emission time, the random part parameters of the sample at the pulse emission time and the parameters of the reconstruction network are updated.

8. A range Doppler domain reconstruction device, characterized in that: include: A training module, used to generate a non-uniform echo signal sample according to a pulse emission time sample, obtain a reconstructed image according to the non-uniform echo signal sample and a preset reconstruction network, construct a loss function according to the pulse emission time sample and the reconstructed image, adjust the parameters of the pulse emission time sample and the reconstruction network according to the loss function, and repeat the training steps until a preset convergence condition is reached; An acquisition module, used for performing pulse emission based on a predetermined pulse emission time to acquire a non-uniform echo signal; The reconstruction module is used to input the non-uniform echo signal into a pre-trained range Doppler domain RD reconstruction network to obtain an RD reconstruction result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the range Doppler domain reconstruction method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the range Doppler domain reconstruction method according to any one of claims 1 to 7 is implemented.

Images