Radar signal processing method

Abstract

The invention relates to a radar signal processing method. The method comprises the following steps: establishing a nonlinear STAP spectrum estimation network and a nonlinear pulse compression network; performing two-dimensional discrete Fourier transform on the digital signal to obtain a Fourier space-time clutter spectrum; inputting the Fourier space-time clutter spectrum into a nonlinear STAP spectrum estimation network to obtain a high-resolution clutter spectrum; reconstructing a CCM based on the high-resolution clutter spectrum, and constructing an STAP filter to realize clutter suppression; and carrying out adaptive pulse compression processing on the signal after clutter suppression through a nonlinear pulse compression network to obtain an adaptive pulse compression result. Compared with the prior art, the method has the advantages that the calculation complexity is reduced, and the key problems of Doppler compensation, sample dependence and the like are solved, so that the radar system shows remarkably stronger robustness for actual environmental factors such as array errors and non-uniform clutters on the premise of keeping the real-time performance.

Description

Technical Field

[0001] The present application relates to the technical field of radar signal processing, and in particular to a radar signal processing method. Background Art

[0002] Existing pulse compression technologies address the trade-off between detection range and resolution through matched filtering or adaptive pulse compression (APC). However, traditional matched filtering introduces range sidelobes, which can overwhelm weak targets with the sidelobes of stronger targets. While APC can suppress these sidelobes, it requires inverting the matrix for each range element, resulting in extremely high computational complexity. While fast APC algorithms reduce computational complexity, they sacrifice system degrees of freedom, reducing the ability to detect targets with low signal-to-noise ratios (SNRs) in densely populated target scenarios. Furthermore, the Doppler effect of moving targets further degrades pulse compression performance. Existing compensation methods (such as Doppler-compensated APC) are computationally inefficient and have limited effectiveness against low-SNR targets.

[0003] Space-Time Adaptive Processing (STAP) technology combines space-time filtering to suppress clutter and enhance target detection. However, its performance relies on a large number of independent and identically distributed (IID) training samples to accurately estimate the clutter covariance matrix (CCM). Traditional STAP performance significantly degrades when samples are insufficient. While dimension reduction STAP and sparse recovery STAP reduce sample requirements, their computational complexity remains high and they lack adaptability to inhomogeneous clutter. While deep learning STAP methods can leverage data-driven approaches to reconstruct the CCM, they suffer from poor interpretability and training difficulties, hindering their stable application in practical systems.

[0004] In traditional airborne radar signal processing, pulse compression is typically placed before STAP, resulting in a significant computational burden and significant Doppler mismatch. Some improved approaches place pulse compression afterward to reduce the computational burden and facilitate Doppler compensation. However, these approaches are still based on linear processing and struggle to cope with signal coupling and non-stationary characteristics in complex environments. Furthermore, both linear STAP and pulse compression involve extensive matrix operations, resulting in low computational efficiency and insufficient robustness against inhomogeneous clutter and low-SNR targets, limiting the overall performance of radar systems. Summary of the Invention

[0005] Based on this, it is necessary to provide an efficient and robust radar signal processing method to improve the radar signal processing performance by addressing at least one of the above-mentioned technical problems: high computational complexity, limited sidelobe suppression capability, insufficient Doppler compensation effect, large sample requirement, and poor adaptability to non-uniform environments.

[0006] To solve the above technical problems, the technical solutions of the present application are as follows:

[0007] In a first aspect, a radar signal processing method comprises:

[0008] A nonlinear STAP spectrum estimation network and a nonlinear pulse compression network are established;

[0009] A two-dimensional discrete Fourier transform is performed on a digitized signal to obtain a Fourier space-time clutter spectrum; wherein the digitized signal comprises a digitized transmitted signal and a corresponding digitized echo signal;

[0010] The Fourier space-time clutter spectrum is input into the nonlinear STAP spectrum estimation network to obtain a high-resolution clutter spectrum;

[0011] Based on the high-resolution clutter spectrum, a CCM is reconstructed, and a STAP filter is constructed to achieve clutter suppression;

[0012] Through the nonlinear pulse compression network, an adaptive pulse compression process is performed on the clutter-suppressed signal to obtain an adaptive pulse compression result.

[0013] In a second aspect, an electronic device comprises:

[0014] A memory for storing computer executable instructions or computer programs;

[0015] A processor for executing the computer executable instructions or computer programs stored in the memory to implement the method of the first aspect.

[0016] In a third aspect, a computer program product comprises computer programs or computer executable instructions, which, when executed by a processor, implement the method of the first aspect.

[0017] Compared with the prior art, the beneficial effects of the technical solutions of the present application are:

[0018] The present application proposes a radar signal processing method, which fundamentally reconstructs the traditional radar signal processing flow, adopts a network cascade processing architecture, and establishes a new processing link of "STAP first and pulse compression later". By replacing the original linear matrix inversion operation with a nonlinear neural network, the overall computational complexity is reduced by two orders of magnitude, while solving key problems such as Doppler compensation and sample dependence. The radar system exhibits significantly stronger robustness to actual environmental factors such as array errors and non-uniform clutter while maintaining real-time performance. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The flowchart of the radar processing method of some embodiments of the present application is shown.

[0020] Figure 2 Schematic diagram of the structure of the nonlinear STAP spectrum estimation network in some embodiments of the present application.

[0021] Figure 3 This is a schematic diagram of the structure of a nonlinear pulse compression network in some embodiments of the present application.

[0022] Figure 4 This is a data flow diagram of the feature map rearrangement operation in some embodiments of the present application.

[0023] Figure 5 Schematic diagram of the structure of the radar processing system in some other embodiments of the present application. DETAILED DESCRIPTION

[0024] The terms "first", "second" etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable in appropriate circumstances, and this is merely a way of distinguishing the objects of the same attribute when describing them in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment. The term "determine" widely covers various actions, may include obtaining, calculating, computing, processing, deriving, investigating, searching (for example, searching in a table, a database or other data structure), ascertaining and similar actions, may also include receiving (for example, receiving information), accessing (for example, accessing data in a memory) and similar actions, may also include generating, creating, establishing and similar actions, and parsing, selecting, selecting and similar actions etc. The relevant definitions of other terms will be provided in the following description.

[0025] It should be noted that when an element is considered to be "connected" to another element, it can be directly connected to the other element or connected to the other element through an intervening element. In addition, the "connection" in the following embodiments should be understood as "electrical connection", "communication connection", etc., if there is transmission of electrical signals or data between the connected objects.

[0026] It should be emphasized that the acquisition, transmission, storage, use, and processing of data in the technical solutions of the embodiments of this application comply with the relevant provisions of national laws and regulations.

[0027] In the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned. They should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0028] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0029] In order to better illustrate this embodiment, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product size;

[0030] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0031] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0032] In some embodiments of the present application, a radar signal processing method is provided. Figure 1 , including steps S110 to S510.

[0033] S110, establishing a nonlinear STAP spectrum estimation network and a nonlinear pulse compression network.

[0034] S210 . Perform a two-dimensional discrete Fourier transform on the digitized signal to obtain a Fourier space-time clutter spectrum; wherein the digitized signal includes a digitized transmit signal and a corresponding digitized echo signal.

[0035] S310 , inputting the Fourier space-time clutter spectrum into the nonlinear STAP spectrum estimation network to obtain a high-resolution clutter spectrum.

[0036] S410 , reconstructing a CCM based on the high-resolution clutter spectrum, and constructing a STAP filter to achieve clutter suppression.

[0037] S510 , performing adaptive pulse compression processing on the clutter suppressed signal through the nonlinear pulse compression network to obtain an adaptive pulse compression result.

[0038] Therefore, in view of the defects of the pulse compression technology and the STAP technology in the existing radar signal processing and the deficiencies of the joint processing of the two, a nonlinear joint processing method based on deep learning is adopted, a neural network is used to replace the traditional matrix inversion operation, the overall operation amount is significantly reduced, the overall calculation complexity is reduced by two orders of magnitude, and the problems of the traditional matched filtering and APC method, such as limited distance sidelobe suppression ability, high calculation complexity, and poor Doppler compensation effect, are overcome. Meanwhile, the traditional radar signal processing flow is reconstructed, a new processing link of “STAP first and pulse compression later” is established, the defects of the traditional STAP technology, such as large demand for training samples, low calculation efficiency, and insufficient adaptive ability to non-uniform clutter, are solved, and the problems of heavy operation burden and Doppler mismatch caused by the processing sequence of pulse compression and STAP in the traditional airborne radar signal processing flow are optimized, so that high-precision, high-efficiency, and robust radar signal processing is finally realized.

[0039] It can be understood that the method provided in the above embodiment reflects the design concept of combining data-driven and model-driven, significantly reduces the dependence on IID (independent and identically distributed) training samples, and still maintains stable performance in a non-uniform clutter environment.

[0040] It should be further pointed out that the method provided in the above embodiment is not only applicable to an airborne radar system, but also can be extended to various platforms such as a shipborne radar system and a satellite-borne radar system, and provides an innovative technical route for a new generation of intelligent radar signal processing system.

[0041] In some embodiments of the present application, the method further includes the step S610:

[0042] S610, constant false alarm rate detection is performed on the signal output by the nonlinear pulse compression network, motion target information is extracted, and robust detection and parameter estimation of a radar target in a complex environment are realized.

[0043] In some embodiments of the present application, the nonlinear STAP spectrum estimation network is established based on the sparse recovery principle and the semi-quadratic splitting algorithm; the nonlinear STAP spectrum estimation network includes a plurality of alternately connected data modules and prior modules, a channel attention module, and a hyperparameter module, and refer to Figure 2 :

[0044] Among them,

[0045] The channel attention module is configured to perform feature enhancement on the Fourier space-time clutter spectrum, and output a channel weight vector for highlighting the feature of the clutter main lobe to the first data module.

[0046] The hyperparameter module is configured to perform feature extraction and parameter adjustment on the Fourier space-time clutter spectrum based on a channel decreasing architecture convolution network, and output parameters to the data module or the prior module.

[0047] a data module, preset with mathematical expressions and without trainable parameters, for realizing a closed-form solution of a data subproblem in the semi-quadratic splitting algorithm and passing to a next said prior module;

[0048] a prior module, for processing a prior subproblem based on a dilated convolutional network and passing to a next said data module after a high Gaussian denoising process realized by batch normalization and ReLU activation; wherein a last said prior module outputs the high-resolution clutter spectrum.

[0049] Firstly, the nonlinear STAP spectrum estimation network adopts a modular convolutional neural network architecture, in which each module strictly corresponds to a mathematical subproblem in the traditional iterative algorithm. Each module is formed into an equivalent iterative unit by cascading, in which the alternating processing of data modules and prior modules simulates the traditional iterative optimization process, and the hyperparameter module realizes the dynamic adaptive adjustment of parameters. This design not only retains the nonlinear advantage of the deep learning network, but also maintains the mathematical essence of the traditional optimization algorithm through mathematical model constraints, ensuring the interpretability of the algorithm and achieving performance improvement through data-driven, which is significantly different from the traditional black-box neural network.

[0050] Secondly, the data module is involved, and the preset mathematical expression is:

[0051]

[0052] wherein z n+1 represents the output of the data module; represents the overall operation of the data module; γ F represents the input clutter Fourier spectrum (i.e. the Fourier space-time clutter spectrum); γ n represents the output of the last prior module; ζ n+1 represents the current equivalent iteration hyperparameters output by the hyperparameter module, which are related to the noise level.

[0053] Further, the hyperparameter module is involved, which adopts a convolutional network to realize dynamic adaptive adjustment of parameters and simultaneously outputs the hyperparameters ζ and η required for the corresponding round equivalent iteration in each data module and prior module, and the process can be represented as:

[0054]

[0055] wherein, represents the overall operation process of the hyperparameter module.

[0056] In addition, the prior module is involved, in which a dilated convolutional network is designed, which can improve the non-uniform clutter adaptability by expanding the receptive field, and the processing process of the prior module can be represented as:

[0057]

[0058] wherein, represents the overall operation process of the prior module.

[0059] It should be further noted that, Figure 2 wherein, represents the overall operation of the channel attention module; I0represents the first equivalent iteration; I k-1 represents the kth equivalent iteration.

[0060] In some embodiments of the present application, the channel attention module comprises an SE-Net (Squeeze and Excitation) component, a global pooling component and a fully connected component for processing complex domain space-time spectral features by using complex convolution, and residual connections are used between convolution layers of the SE-Net component.

[0061] The above embodiments retain phase information through a complex domain network structure, ensure effective gradient propagation through residual connection design, greatly reduce the computational complexity of traditional sparse recovery STAP while maintaining algorithm interpretability, and effectively improve the robustness of the system to array errors, Doppler frequency shifts and other non-ideal factors. Illustratively, the channel attention module enhances the features of the clutter spectrum after Fourier transform based on an improved (capable of complex convolution) SE-Net structure, and generates a channel weight vector through a global pooling layer and two fully connected layers.

[0062] In some embodiments of the present application, the convolution operation in the nonlinear STAP spectral estimation network all uses a 3x3 convolution kernel; wherein the number of channels of the channel attention module is 16; the prior module uses a three-layer dilated convolution network; the hyperparameter module uses a five-layer convolution network to optimize the iteration parameters, and the number of channels is 64, 32, 16, 16 and 16 in turn, that is, the channel decreasing architecture of 64→32→16→16→16 is used to realize feature extraction and parameter adjustment.

[0063] It should be noted that using a 3x3 convolution kernel throughout can accelerate the network operation process, while avoiding the complexity of hyperparameters leading to difficulty in reproducing the network.

[0064] In some embodiments of the present application, the nonlinear pulse compression network comprises a plurality of cascaded sub-networks; wherein each of the sub-networks comprises a Doppler compensation branch (referred to as a "Doppler module") and a power estimation branch (referred to as a "power module") arranged in parallel, and realizes feature fusion through cross-level residual connections; the output ends of the Doppler compensation branch and the power estimation branch are respectively connected to a pulse compression branch (referred to as a "pulse compression module").

[0065] To solve the problem of joint processing of pulse compression and STAP, the embodiments of the present application construct an innovative end-to-end processing framework. In the STAP processing link, the complex convolution operation is used to retain the space-time phase information, and the residual connection structure is used to solve the stability problem of complex network training, so as to realize the nonlinear clutter suppression effect that cannot be achieved by traditional methods. In the pulse compression link, the amplitude estimation module and the Doppler shift estimation module are set in parallel to realize parameter decoupling, and the sub-network stacking structure is used to simulate the traditional iterative optimization process, thereby breaking through the performance limit of linear adaptive pulse compression.

[0066] Figure 3 The network framework of the nonlinear pulse compression network is shown, which takes the received echo signal and the transmitted signal as double input sources, synchronously extracts the signal feature parameters of each distance unit through the parallel power module and the Doppler module, then takes the outputs of the two modules as the inputs of the pulse compression module, and finally generates the optimized adaptive pulse compression result. Each processing unit (including the power module, the Doppler module and the pulse compression module) is regarded as a basic sub-network, and the equivalent iterative optimization is realized through the cascade stacking of multiple sub-networks, so that the progressive performance improvement is obtained while the network structure simplicity is maintained. This architecture not only ensures the specialized processing capability of each functional module, but also enhances the parameter estimation accuracy and pulse compression effect of the system through the iterative mechanism.

[0067] In some embodiments of the present application, the Doppler compensation branch uses dilated convolution to realize Doppler frequency shift feature extraction, and performs feature map rearrangement operation on the input data in the convolution process to equivalent time-frequency transformation to obtain the rearranged feature map for subsequent dilated convolution, and then inversely rearranges and deconvolves to a one-dimensional signal after completing feature extraction; and the power estimation branch uses a multi-layer perceptron based on a dilated convolution kernel.

[0068] Thus, the nonlinear filtering characteristics of the deep convolutional neural network are used to replace the traditional matched filtering, which suppresses the distance sidelobes while avoiding the performance loss caused by Doppler mismatch.

[0069] As a non-limiting example, the Doppler compensation branch uses two to three layers of dilated convolution to realize.

[0070] In some examples of the present application, the pulse compression branch is composed of a CNN network based on an encoder-decoder structure, which aims to generate a pulse compression echo intermediate signal or an adaptive pulse compression result based on Doppler shift estimation features, amplitude estimation features and radar transmission signals.

[0071] For the feature map rearrangement operation, Figure 4 A data flow diagram of the feature map rearrangement operation in the Doppler compensation branch is given. First, the input data with a length of (2N-1) is formed (KN C) individual channels, then the feature map of each channel is rearranged into a two-dimensional matrix to be equivalent to a multi-group time-frequency transform, which can effectively extract the Doppler frequency shift characteristics in different time delays in the echo signal, and finally inversely rearranged and deconvoluted into a one-dimensional signal with a length of (2N-1).

[0072] It should be noted that in the embodiments of the present application, the modular network architecture has high scalability, and each functional module can be independently optimized and upgraded. Compared with the traditional STAP system which requires special hardware acceleration, the method provided in the present application can be efficiently deployed on a general computing platform, greatly reducing the system implementation cost.

[0073] In some embodiments of the present application, in step S110, the nonlinear STAP spectrum estimation network optimizes the network parameters through end-to-end training, and the training process includes steps S1110 to S1120.

[0074] S1110, based on the radar system parameters and the radar range equation, an echo signal model is constructed, and simulation modeling is performed considering various non-ideal factors to obtain a training data set; the training data set includes distance gate data to be detected and corresponding transmit signal data under various non-ideal conditions.

[0075] Through simulation modeling, the generated simulation data will reflect different signal-to-noise ratios, Doppler frequency shifts and clutter distribution characteristics.

[0076] S1120, the nonlinear STAP spectrum estimation network and the nonlinear pulse compression network are trained on the training data set, and the network parameters are updated until the training is completed.

[0077] In some embodiments of the present application, the offline training process of the nonlinear pulse compression network includes steps S1111 to S1112.

[0078] S1111, based on the radar system parameters and the radar range equation, an echo signal model is constructed, and simulation is performed considering various non-ideal factors to obtain distance gate simulation data to be detected and construct a training data set with corresponding simulation transmit signals; wherein the label in the training data set is constructed using the following formula:

[0079] GT = n + gt

[0080] Wherein, n represents a noise vector;

[0081]

[0082] In the formula, GT represents the true value of the label in the training data set; T represents a distance unit set containing a target; n represents the index of the distance unit, i.e. the nth distance unit.

[0083] S1112, let the non-linear pulse compression network train on the training data set, and update the network parameters based on the loss function until the training is completed; wherein the loss function is as follows:

[0084]

[0085] In the formula, GT represents the label true value in the training data set; represents the prediction value of the network output; W n represents a penalty coefficient, and

[0086]

[0087] Wherein, T represents a distance unit set containing a target.

[0088] Some embodiments of the application also provide a whole process example from offline training to online learning to facilitate implementation by those skilled in the art. In this example, a radar signal processing system is constructed, such as Figure 5 , including an offline training framework and an online learning framework; wherein the offline training framework includes a radar echo modeling unit 1, a training data set construction unit 2, a neural network construction unit 3, a neural network training unit 4, the online learning framework includes a radar received echo data 5, an analog-digital conversion unit 6, a Fourier spectrum estimation unit 7, a non-linear STAP spectrum estimation unit 8, a STAP processing unit 9, a non-linear pulse compression unit 10, a constant false alarm detection processing unit 11, and the system execution method steps include steps S100 to S1100:

[0089] S100, the radar echo modeling unit 1 constructs an echo signal model based on the radar system parameters and the radar distance equation, and considers various non-ideal factors to simulate and model the radar echo signal under complex clutter scenes, and generates training data containing different signal-to-noise ratios, Doppler shifts and clutter distribution characteristics; wherein the spatial error considers that the array element amplitude and phase error satisfies the complex Gaussian distribution, and the variance is selected from 0 to 0.05 with an interval of 0.002;

[0090] S200, in the training data set construction unit 2, the simulated distance gate data to be detected under various non-ideal conditions and the transmitted signal are taken as the training set;

[0091] S300, in the constructed neural network unit 3, the constructed nonlinear STAP spectrum estimation network adopts a (3x3) convolution kernel, wherein the attention module is 16 channels; the channel numbers of the three convolution layers of the prior module are 32, 16 and 1 respectively; the channel numbers of the super parameter module are 32, 16, 16 and 16 respectively. The convolution kernel sizes of the two expansion convolution layers of the power module of the nonlinear pulse compression network are (1x9) and (1x3) respectively; the convolution kernel sizes of the two expansion convolution layers of the Doppler module are (1x18) and (1x3) respectively; the convolution kernel sizes of the three convolution layers of the pulse pressure module are (1x9), (1x3) and (1x1) respectively;

[0092] S400, in the training neural network unit 4, each network is trained based on the training data set, the batch size is set to 16, and all samples are trained for 100 times;

[0093] S500, in the radar received echo data unit 5, N elements K pulses of signals are received in each range gate;

[0094] S600, in the analog-digital conversion unit 6, the received echo signal and the corresponding transmission signal are analog-digital converted, the received data and the transmission signal are digitized and stored;

[0095] S700, in the Fourier spectrum estimation 7, the digitized signal stored in step S600 is selected to have a suitable length for Fourier spectrum estimation;

[0096] S800, in the nonlinear STAP spectrum estimation unit 8, the offline trained nonlinear STAP spectrum estimation network is deployed, and the high resolution clutter spectrum is obtained by using the Fourier spectrum;

[0097] S900, in the STAP processing unit 9, the network output spectrum of step S800 is used to reconstruct CCM and construct STAP filter;

[0098] S1000, in the nonlinear pulse compression unit 10, the offline trained nonlinear pulse compression network is deployed, and the STAP processed data of step S900 is adaptively pulse compressed;

[0099] S1100, the constant false alarm detection processing unit 11 performs constant false alarm processing on the adaptively pulse compressed data, and completes the detection processing of the radar moving target.

[0100] It should be understood that the above-mentioned echo modeling, data set construction, neural network construction, neural network training, analog-digital conversion, neural network processing and constant false alarm detection processing can be realized on a general programmable signal processing board.

[0101] Some embodiments of the present application also provide a computer readable storage medium, on which at least one instruction, at least one program, a code set or an instruction set is stored, the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor, so that the processor executes part or all steps of the method provided in the foregoing embodiments.

[0102] It can be understood that the storage medium can be transitory or non-transitory. Illustratively, the storage medium includes, but is not limited to, a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0103] Illustratively, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0104] Illustratively, the read-only memory includes, but is not limited to, a mask ROM, a PROM, an EPROM, an EEPROM, a Flash, etc.

[0105] Illustratively, the random access memory includes, but is not limited to, a DRAM, an SRAM, an SDRAM, a DDR SDRAM, etc.

[0106] In some examples, a computer program product is provided, which can be implemented by hardware, software or a combination thereof. As a non-limiting example, the computer program product can be embodied as the storage medium, and can also be embodied as a software product, such as an SDK (Software Development Kit) or the like.

[0107] As a non-limiting example, a computer program product is provided, which includes a computer program or computer executable instructions stored in a computer readable storage medium. The processor of the electronic device reads the computer program or computer executable instructions from the computer readable storage medium, and the processor executes the computer executable instructions, so that the electronic device executes part or all steps of the method provided in the embodiments of the present application.

[0108] In some examples, a computer program is provided, including computer readable code which, when run in a computer device, causes a processor in the computer device to perform some or all of the steps of the methods.

[0109] In some embodiments of the present application, an electronic device is also provided, including a memory and a processor, the memory storing at least one instruction, at least one program, a code set or an instruction set, and the processor implementing some or all of the steps of the method as described in the foregoing embodiments when executing the at least one instruction, at least one program, code set or instruction set.

[0110] In some examples, a hardware entity of the electronic device is provided, including a processor, a memory and a communication interface; wherein the processor generally controls the overall operation of the electronic device; the communication interface is used for the electronic device to communicate with other terminals or servers through a network; the memory is configured to store instructions and applications executable by the processor, and can also cache data to be processed by the processor and data to be processed or having been processed by each module in the electronic device (including but not limited to image data, audio data, voice communication data and video communication data), which can be realized by FLASH, EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory) or RAM (Random Access Memory).

[0111] The processor can include one or more processing elements. Thus, the processor can include one or more integrated circuits (ICs) that are configured to perform the functions of the processor. In addition, each integrated circuit can include circuitry (e.g., first circuitry, second circuitry, and other circuitry) that is configured to perform the functions of the processor.

[0112] Further, the processor, the communication interface and the memory can transmit data through a bus, which can include any number of interconnected buses and bridges, connecting the various circuits of the one or more processors and the memory together.

[0113] The same or similar reference numerals correspond to the same or similar components;

[0114] The terms describing the positional relationship in the drawings are only used for exemplary illustration, and cannot be understood as a limitation on the present application;

[0115] It should be noted that the embodiments and the features in the embodiments in the present application can be combined with each other without conflict.

[0116] In different specific implementations, the methods or systems described in the present application can be implemented in software, hardware or a combination thereof. In addition, the order of the steps of the method can be changed, and various elements can be added, reordered, combined, omitted, modified, etc.

[0117] Obviously, the above embodiments of the present application are only examples for clearly illustrating the present application, and are not intended to limit the implementation manners of the present application, and are not used to limit the present application. Based on the above description, other different forms of changes or variations can be made by those skilled in the art, and each discrete structure / function module or unit can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part, and the structure and function of the discrete components can be implemented as a combined structure or component. Here, it is not necessary and also impossible to enumerate all the implementation manners. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the claims of the present application.

Claims

1. A radar signal processing method, characterized in that: include: Establish nonlinear STAP spectrum estimation network and nonlinear pulse compression network; Performing a two-dimensional discrete Fourier transform on the digitized signal to obtain a Fourier space-time clutter spectrum; wherein the digitized signal includes a digitized transmit signal and a corresponding digitized echo signal; Inputting the Fourier space-time clutter spectrum into the nonlinear STAP spectrum estimation network to obtain a high-resolution clutter spectrum; Reconstructing CCM based on the high-resolution clutter spectrum and constructing a STAP filter to achieve clutter suppression; The nonlinear pulse compression network is used to perform adaptive pulse compression processing on the signal after clutter suppression to obtain an adaptive pulse compression result.

2. A radar signal processing method according to claim 1, characterized in that: The nonlinear STAP spectrum estimation network is established based on the sparse recovery principle and the semi-quadratic splitting algorithm; The nonlinear STAP spectrum estimation network includes several alternately connected data modules and prior modules, as well as a channel attention module and a hyperparameter module: a channel attention module, configured to perform feature enhancement on the Fourier space-time clutter spectrum and output a channel weight vector for highlighting the clutter main lobe feature to the first data module; A hyperparameter module, configured to extract features and adjust parameters of the Fourier space-time clutter spectrum based on a convolutional network with a channel-decreasing architecture, and output the parameters to the data module or the prior module; A data module, which is preset with a mathematical expression and does not contain trainable parameters, is used to implement a closed-form solution to the data subproblem in the semi-quadratic splitting algorithm and pass it to the next prior module; The priori module is used to process the priori sub-problem based on the dilated convolutional network, and pass it to the next data module after implementing the Gaussian noise reduction process through batch normalization and ReLU activation; wherein the last priori module outputs the high-resolution clutter spectrum.

3. The radar signal processing method according to claim 2, wherein: The channel attention module includes an SE-Net component that uses complex convolution to process complex domain spatiotemporal features, a global pooling component and a fully connected component, and residual connections are used between the convolution layers of the SE-Net component.

4. The radar signal processing method according to claim 2, wherein: The convolution operations in the nonlinear STAP spectral estimation network all use 3×3 convolution kernels; the number of channels of the channel attention module is 16; the prior module uses a three-layer dilated convolutional network; the hyperparameter module uses a five-layer convolutional network, and the number of channels is 64, 32, 16, 16 and 16 respectively.

5. The radar signal processing method according to claim 2, wherein: The nonlinear pulse compression network includes multiple cascaded sub-networks; wherein each of the sub-networks includes a Doppler compensation branch and a power estimation branch arranged in parallel, and realizes feature fusion through cross-level residual connection; the output ends of the Doppler compensation branch and the power estimation branch are respectively connected to the pulse compression branch.

6. A radar signal processing method according to claim 5, characterized in that: The nonlinear STAP spectrum estimation network optimizes network parameters through end-to-end training, and the training process includes: An echo signal model is constructed based on radar system parameters and radar range equations, and simulation modeling is performed considering various non-ideal factors to obtain a training data set; the training data set includes various range gate data to be detected under non-ideal conditions and corresponding transmission signal data; The nonlinear STAP spectrum estimation network is trained on the training data set, and the network parameters are updated until the training is completed.

7. The radar signal processing method according to claim 5, characterized in that: The Doppler compensation branch uses dilated convolution to extract Doppler frequency shift features, and performs a feature map rearrangement operation on the input data during the convolution process to achieve an equivalent time-frequency transformation to obtain a rearranged feature map for subsequent dilated convolution. After feature extraction is completed, the feature map is reversely rearranged and deconvolved into a one-dimensional signal. In addition, the power estimation branch is implemented using a multi-layer perceptron based on a dilated convolution kernel.

8. A radar signal processing method according to any one of claims 1 to 7, characterized in that: Also includes: Constant false alarm detection is performed on the signal output by the nonlinear pulse compression network to extract moving target information, thereby achieving robust detection and parameter estimation of radar targets in complex environments.

9. An electronic device, characterized in that: include: a memory for storing computer executable instructions or computer programs; The processor is configured to implement the method according to any one of claims 1 to 9 when executing the computer-executable instructions or computer program stored in the memory.

10. A computer program product comprising a computer program or computer executable instructions, characterized in that When the computer program or computer executable instructions are executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images