Radar target feature recognition method based on artificial intelligence

Abstract

The invention discloses a radar target feature recognition method based on artificial intelligence, and the method comprises the steps: carrying out the analog-to-digital conversion of a received echo signal, and obtaining the time domain sampling data of the echo signal; the time domain sampling data is preprocessed, a candidate signal sequence is obtained, and preprocessing comprises at least one of pulse compression, moving target detection and clutter suppression; time-frequency transformation and multidimensional feature calculation are carried out on the candidate signal sequence, multidimensional features of the candidate signal sequence are extracted, and the multidimensional features comprise an amplitude feature, a phase feature, a frequency spectrum feature and a kinematics feature; the multi-dimensional features are constructed into feature vectors according to a preset format, and the feature vectors comprise a time domain feature component, a frequency domain feature component and a space domain feature component; and inputting the feature vector into a target recognition model which is trained by adopting a convolutional neural network (CNN) and a recurrent neural network (RNN) in advance, and outputting a target feature recognition result.

Description

Technical Field

[0001] This invention relates to the field of optoelectronic technology, and in particular to a radar target feature recognition method based on artificial intelligence. Background Technology

[0002] With the development of electronic countermeasures and radar detection technologies, radar target feature recognition plays a crucial role in both military and civilian applications. Traditional direction finding and recognition methods mainly include multibeam amplitude comparison direction finding and interferometric direction finding. Multibeam amplitude comparison direction finding determines the target's angle of arrival by comparing the relative amplitudes of signals received from adjacent beams, but it places strict requirements on beam design, and sidelobes are prone to causing nonlinear errors. Interferometric direction finding calculates the angle of arrival by the phase difference between two received signals, and while structurally simple, it is easily limited by phase ambiguity and image ambiguity. Although these methods have been improved, they are mostly limited to azimuth measurement and struggle to simultaneously achieve spectral feature recognition, exhibiting shortcomings in complex electromagnetic environments.

[0003] Chinese invention patent CN114325628A discloses "A method and apparatus for radar target feature recognition based on optical delay fast interferometric scanning." This method utilizes an electro-optic converter and an optical delay numerically controlled scanning device to control the delay of the received signal. Signal synthesis is achieved through an optical combiner and photoelectric detection, and the radio frequency signal power is calculated by a power characteristic analysis device. By analyzing the correspondence between the peak power and the delay amount through forward and reverse optical delay scanning, the azimuth angle, frequency, and spectral characteristics of the target signal are measured. This method is applicable to radar target detection under different electromagnetic environments and modulation schemes.

[0004] However, the aforementioned existing technologies still rely on optical delay hardware and power detection to achieve target recognition, and suffer from problems such as signal processing dependence on physical devices, limited feature extraction methods, and difficulty in adapting to diverse and complex signal environments. Summary of the Invention

[0005] To address the technical problems of existing technologies that rely on optical delay hardware and power detection for target recognition, have limitations in signal processing due to physical devices, employ limited feature extraction methods, and struggle to adapt to diverse and complex signal environments, this invention provides an artificial intelligence-based radar target feature recognition method.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a radar target feature recognition method based on artificial intelligence, comprising:

[0008] The received echo signal is converted from analog to digital to obtain the time-domain sampling data of the echo signal;

[0009] The time-domain sampled data is preprocessed to obtain a candidate signal sequence, wherein the preprocessing includes at least one of pulse compression, moving target detection, and clutter suppression;

[0010] The candidate signal sequence is subjected to time-frequency transformation and multi-dimensional feature calculation to extract multi-dimensional features of the candidate signal sequence, wherein the multi-dimensional features include amplitude features, phase features, spectral features and kinematic features;

[0011] The multidimensional features are constructed into a feature vector according to a preset format, wherein the feature vector includes time-domain feature components, frequency-domain feature components, and spatial-domain feature components;

[0012] The feature vector is input into a target recognition model that has been pre-trained using a convolutional neural network (CNN) and a recurrent neural network (RNN), and the target feature recognition result is output.

[0013] In one alternative embodiment, the preprocessing includes clutter suppression;

[0014] The preprocessing of the time-domain sampled data to obtain the candidate signal sequence includes:

[0015] The STAP algorithm, a space-time adaptive processing algorithm, is used to suppress fixed clutter in the time-domain sampled data to obtain a clutter-suppressed signal sequence.

[0016] Candidate signal sequences are obtained based on the clutter-suppressed signal sequence.

[0017] In one optional embodiment, the weight vector of the STAP algorithm is iteratively calculated using the least mean square (LMS) update method, and the iterative calculation converges successively based on the covariance matrix of the signal sequence.

[0018] In one optional embodiment, the time-frequency transformation of the candidate signal sequence includes:

[0019] The candidate signal sequence is divided into frames according to a preset window length, and each frame signal is multiplied with a preset weighted window function to obtain a windowed signal.

[0020] Perform a Fast Fourier Transform (FFT) on the windowed signal to obtain the spectral distribution of each signal;

[0021] The spectral distributions of each frame are spliced ​​together in chronological order to form a time-frequency matrix, and local instantaneous frequency features are extracted from the time-frequency matrix.

[0022] In one optional embodiment, the multidimensional feature calculation of the candidate signal sequence includes:

[0023] The candidate signal sequence is convolved with a preset mother wavelet function to obtain wavelet coefficients at different scales.

[0024] The wavelet energy distribution of the wavelet coefficients at multiple preset scales is calculated, and the wavelet energy distribution is used to characterize the local characteristics of the signal in the time domain and frequency domain.

[0025] The wavelet energy distribution is arranged according to the scale factor to construct time-frequency feature maps at multiple preset scales, thereby obtaining the spectral features of the candidate signal sequence.

[0026] In one optional embodiment, the multidimensional feature calculation of the candidate signal sequence includes:

[0027] Perform absolute value operation on the candidate signal sequence to obtain the amplitude sequence;

[0028] Envelope detection is performed on the amplitude sequence to obtain the amplitude envelope;

[0029] Statistical parameters are calculated based on the amplitude envelope and used as amplitude features of the candidate signal sequence.

[0030] In one optional embodiment, the multidimensional feature calculation of the candidate signal sequence includes:

[0031] Perform a Hilbert transform on the candidate signal sequence to obtain the analytic signal;

[0032] The instantaneous phase is calculated based on the real and imaginary parts of the analyzed signal;

[0033] The instantaneous phase is expanded to form a continuous phase sequence, which serves as the phase feature of the candidate signal sequence.

[0034] In one optional embodiment, the multidimensional feature calculation of the candidate signal sequence includes:

[0035] The Doppler frequency shift is calculated based on the candidate signal sequence at multiple time points to obtain the target velocity;

[0036] The target acceleration is obtained by performing a first-order difference calculation on the target velocity;

[0037] Based on the time series of the target velocity and the target acceleration, track stability parameters are calculated as kinematic features of the candidate signal sequence.

[0038] In one optional embodiment, constructing the multidimensional features into a feature vector according to a preset format includes:

[0039] The amplitude feature, phase feature, spectral feature, and kinematic feature are normalized to obtain a normalized feature sequence;

[0040] According to the preset feature mapping relationship, the normalized feature sequence is divided into time domain feature components, frequency domain feature components and spatial domain feature components.

[0041] The time-domain feature components, the frequency-domain feature components, and the spatial-domain feature components are concatenated in a preset order to form a feature vector of uniform length.

[0042] In an optional embodiment, the step of inputting the feature vector into a target recognition model pre-trained with a convolutional neural network and a recurrent neural network, and outputting the target feature recognition result, includes:

[0043] The feature vector is input into the input layer of the CNN, and local spatial features are extracted through convolution and pooling operations.

[0044] The local spatial features are expanded and input into the RNN, and sequence dependency features are obtained through time series modeling.

[0045] The sequence-dependent features are input into the fully connected layer and the classification layer, and the corresponding target category label is output to obtain the target feature recognition result.

[0046] The beneficial effects of the technical solution provided by this invention include at least the following:

[0047] By employing a step-by-step chain of analog-to-digital conversion, preprocessing, time-frequency transformation, and multi-dimensional feature extraction, more comprehensive target information can be obtained in complex electromagnetic environments. The introduction of multiple features, including amplitude, phase, spectrum, and kinematics, ensures the multidimensionality of target feature description and improves feature discriminability. A feature vector construction mechanism unifies the encoding of different types of feature components, facilitating direct processing by artificial intelligence models. The combination of convolutional neural networks and recurrent neural networks enables the extraction of local pattern features and the capture of temporal dependencies. The overall method, centered on a step-by-step processing chain, can be implemented on general-purpose radar platforms without relying on complex optical or special hardware devices, exhibiting good adaptability and scalability. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1This is a flowchart illustrating a radar target feature recognition method based on artificial intelligence provided in an embodiment of the present invention. Detailed Implementation

[0050] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The following embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes, but the scope of protection of the present invention is not limited to the following embodiments.

[0051] In this invention, the terms "in one possible embodiment," "exemplary," or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "in one possible embodiment," "exemplary," or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "in one possible embodiment," "exemplary," or "for example" is intended to present the relevant concepts in a specific manner.

[0052] Figure 1 This is a flowchart illustrating an artificial intelligence-based radar target feature recognition method provided in an embodiment of the present invention. Figure 1 As shown, an artificial intelligence-based radar target feature recognition method includes:

[0053] S101, the received echo signal is converted from analog to digital to obtain the time-domain sampling data of the echo signal.

[0054] S102, the time-domain sampled data is preprocessed to obtain a candidate signal sequence.

[0055] The preprocessing includes at least one of pulse compression, moving target detection, and clutter suppression.

[0056] In one optional embodiment, the preprocessing includes clutter suppression;

[0057] The preprocessing of the time-domain sampled data to obtain the candidate signal sequence includes:

[0058] The STAP algorithm, a space-time adaptive processing algorithm, is used to suppress fixed clutter in the time-domain sampled data to obtain a clutter-suppressed signal sequence.

[0059] Candidate signal sequences are obtained based on the clutter-suppressed signal sequence.

[0060] For example, the STAP algorithm is applied to suppress ground clutter by airborne radar. Time-domain sampled data is input into a two-dimensional filter matrix. A 32-point delay unit is used in the time domain, and an 8-element array antenna is used in the spatial domain. The filter weights are calculated and updated by the STAP algorithm, and the final suppression ratio can reach 40dB, effectively extracting candidate signal sequences of moving targets.

[0061] In one optional embodiment, the weight vector of the STAP algorithm is iteratively calculated using the least mean square (LMS) update method, and the iterative calculation converges successively based on the covariance matrix of the signal sequence.

[0062] For example, the LMS update method sets the step size parameter μ = 0.01, the initial weight vector is set to zero, and each update uses the latest estimate of the covariance matrix for iteration. After 100 iterations, the filter weights gradually converge, and the output signal convergence error is less than -25dB.

[0063] S103, perform time-frequency transformation and multi-dimensional feature calculation on the candidate signal sequence to extract the multi-dimensional features of the candidate signal sequence.

[0064] The multidimensional features include amplitude features, phase features, spectral features, and kinematic features.

[0065] In one optional embodiment, the time-frequency transformation of the candidate signal sequence includes:

[0066] The candidate signal sequence is divided into frames according to a preset window length, and each frame signal is multiplied with a preset weighted window function to obtain a windowed signal.

[0067] Perform a Fast Fourier Transform (FFT) on the windowed signal to obtain the spectral distribution of each signal;

[0068] The spectral distributions of each frame are spliced ​​together in chronological order to form a time-frequency matrix, and local instantaneous frequency features are extracted from the time-frequency matrix.

[0069] For example, the STFT uses a Hamming window with a length of 256 points and a window spacing of 128 points. The spectral distribution is obtained by performing an FFT on each windowed signal. The resulting spectral matrix is ​​visualized as a time-frequency image for further extraction of local instantaneous frequency features, such as Doppler shift patterns.

[0070] In one optional embodiment, the multidimensional feature calculation of the candidate signal sequence includes:

[0071] The candidate signal sequence is convolved with a preset mother wavelet function to obtain wavelet coefficients at different scales.

[0072] The wavelet energy distribution of the wavelet coefficients at multiple preset scales is calculated, and the wavelet energy distribution is used to characterize the local characteristics of the signal in the time domain and frequency domain.

[0073] The wavelet energy distribution is arranged according to the scale factor to construct time-frequency feature maps at multiple preset scales, thereby obtaining the spectral features of the candidate signal sequence.

[0074] For example, the Morlet wavelet is used as the mother wavelet function. When convolving the candidate signal sequence, the scale factor is set to a logarithmic sequence from 2 to 64. After calculating the energy distribution of the obtained wavelet coefficients, a time-frequency feature map is plotted, and the main frequency band energy of the signal is captured in the scale range of 20–40 as the spectral feature input.

[0075] In one optional embodiment, the multidimensional feature calculation of the candidate signal sequence includes:

[0076] Perform absolute value operation on the candidate signal sequence to obtain the amplitude sequence;

[0077] Envelope detection is performed on the amplitude sequence to obtain the amplitude envelope;

[0078] Statistical parameters are calculated based on the amplitude envelope and used as amplitude features of the candidate signal sequence.

[0079] For example, during amplitude feature extraction, the absolute values ​​of the candidate signals are first taken to obtain an amplitude sequence, and then a smooth envelope is generated by averaging through a sliding window (window length 16 points). The mean, variance, and peak value of the envelope are calculated as the amplitude features of that segment of the signal.

[0080] In one optional embodiment, the multidimensional feature calculation of the candidate signal sequence includes:

[0081] Perform a Hilbert transform on the candidate signal sequence to obtain the analytic signal;

[0082] The instantaneous phase is calculated based on the real and imaginary parts of the analyzed signal;

[0083] The instantaneous phase is expanded to form a continuous phase sequence, which serves as the phase feature of the candidate signal sequence.

[0084] For example, during phase feature extraction, the Hilbert transform is used to obtain the analytic signal, and the instantaneous phase is calculated for the real and imaginary parts respectively. The instantaneous phase unwrapping uses the "unwrap" algorithm to eliminate π jumps, resulting in a continuous phase sequence, and the linear phase slope is extracted in the range of 0–100 μs as the feature input.

[0085] In one optional embodiment, the multidimensional feature calculation of the candidate signal sequence includes:

[0086] The Doppler frequency shift is calculated based on the candidate signal sequence at multiple time points to obtain the target velocity;

[0087] The target acceleration is obtained by performing a first-order difference calculation on the target velocity;

[0088] Based on the time series of the target velocity and the target acceleration, track stability parameters are calculated as kinematic features of the candidate signal sequence.

[0089] For example, during kinematic feature extraction, the Doppler frequency shift is calculated using candidate signals from multiple time slices, and the target velocity is obtained using the formula Δf = fdoppler. Then, the velocity curve is subjected to first-order difference to obtain the acceleration curve. Combining the variance indices of velocity and acceleration, trajectory stability parameters are calculated to reflect the target's maneuverability.

[0090] S104, construct the multidimensional features into a feature vector according to a preset format.

[0091] The feature vector includes time-domain feature components, frequency-domain feature components, and spatial-domain feature components.

[0092] In one optional embodiment, constructing the multidimensional features into a feature vector according to a preset format includes:

[0093] The amplitude feature, phase feature, spectral feature, and kinematic feature are normalized to obtain a normalized feature sequence;

[0094] According to the preset feature mapping relationship, the normalized feature sequence is divided into time domain feature components, frequency domain feature components and spatial domain feature components.

[0095] The time-domain feature components, the frequency-domain feature components, and the spatial-domain feature components are concatenated in a preset order to form a feature vector of uniform length.

[0096] For example, in one embodiment, when constructing the feature vector, the amplitude, phase, spectrum, and kinematic features are normalized to the [0,1] interval. According to the mapping relationship, the time-domain component has a length of 128, the frequency-domain component has a length of 256, and the spatial-domain component has a length of 64, which are finally concatenated into a 448-dimensional feature vector, which is used as the input of the neural network.

[0097] S105, the feature vector is input into the target recognition model that has been pre-trained using a convolutional neural network (CNN) and a recurrent neural network (RNN), and the target feature recognition result is output.

[0098] In an optional embodiment, the step of inputting the feature vector into a target recognition model pre-trained with a convolutional neural network and a recurrent neural network, and outputting the target feature recognition result, includes:

[0099] The feature vector is input into the input layer of the CNN, and local spatial features are extracted through convolution and pooling operations.

[0100] The local spatial features are expanded and input into the RNN, and sequence dependency features are obtained through time series modeling.

[0101] The sequence-dependent features are input into the fully connected layer and the classification layer, and the corresponding target category label is output to obtain the target feature recognition result.

[0102] For example, the pre-trained model can be trained using transfer learning. First, CNN weights pre-trained on the ImageNet dataset are loaded as initialization, and then fine-tuned using a radar target dataset. This can accelerate convergence and improve the ability to recognize a small number of samples.

[0103] For example, the system supports federated learning. Different radar platforms independently train the model's parameter update gradients, and then upload the encrypted gradients to a central server for aggregation, thereby achieving cross-platform model updates and ensuring data privacy.

[0104] For example, the feature extraction and neural network computing modules are deployed on a heterogeneous hardware platform combining FPGA and GPU. The STFT and CWT feature extraction parts are implemented through a parallel pipeline on the FPGA, while the CNN and RNN models are accelerated by the GPU, with the overall recognition latency controlled within 10ms.

[0105] In one embodiment, the radar system transmits at the X-band (8–12 GHz) with a sampling rate of 200 MHz. Echo signals are acquired via a high-speed analog-to-digital converter to form time-domain sampled data. Preprocessing employs a 64-point pulse compression matched filter, and moving target detection uses the constant false alarm rate (CFAR) algorithm. The time-frequency transformation part combines STFT and CWT to extract multi-scale features, ultimately forming a feature vector containing 128-dimensional time-domain components, 256-dimensional frequency-domain components, and 64-dimensional spatial-domain components, which is then input into a deep learning model for classification.

[0106] Based on the above method, this invention, through a step-by-step chain of analog-to-digital conversion, preprocessing, time-frequency transformation, and multi-dimensional feature extraction, can obtain more comprehensive target information in complex electromagnetic environments. It introduces multiple types of features, such as amplitude, phase, spectrum, and kinematics, ensuring the multidimensionality of target feature description and improving feature discriminability. Utilizing a feature vector construction mechanism, it uniformly encodes different types of feature components, facilitating direct processing by artificial intelligence models. Employing a structure combining convolutional neural networks and recurrent neural networks, it can extract local pattern features and capture temporal dependencies. The overall method, with a step-by-step processing chain at its core, can be implemented on a general-purpose radar platform without relying on complex optical or special hardware devices, exhibiting good adaptability and scalability.

[0107] Furthermore, it should be noted that the present invention can be provided as an apparatus or a computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0108] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0109] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0110] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0111] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A radar target feature recognition method based on artificial intelligence, characterized in that, include: The received echo signal is converted from analog to digital to obtain the time-domain sampling data of the echo signal; The time-domain sampled data is preprocessed to obtain a candidate signal sequence, wherein the preprocessing includes at least one of pulse compression, moving target detection, and clutter suppression; The candidate signal sequence is subjected to time-frequency transformation and multi-dimensional feature calculation to extract multi-dimensional features of the candidate signal sequence, wherein the multi-dimensional features include amplitude features, phase features, spectral features and kinematic features; The multidimensional features are constructed into a feature vector according to a preset format, wherein the feature vector includes time-domain feature components, frequency-domain feature components, and spatial-domain feature components; The feature vector is input into a target recognition model that has been pre-trained using a convolutional neural network (CNN) and a recurrent neural network (RNN), and the target feature recognition result is output.

2. The radar target feature recognition method based on artificial intelligence as described in claim 1, characterized in that, The preprocessing includes clutter suppression; The preprocessing of the time-domain sampled data to obtain the candidate signal sequence includes: The STAP algorithm, a space-time adaptive processing algorithm, is used to suppress fixed clutter in the time-domain sampled data to obtain a clutter-suppressed signal sequence. Candidate signal sequences are obtained based on the clutter-suppressed signal sequence.

3. The radar target feature recognition method based on artificial intelligence as described in claim 2, characterized in that, The weight vector of the STAP algorithm is iteratively calculated using the least mean square (LMS) update method, and the iterative calculation converges successively based on the covariance matrix of the signal sequence.

4. The radar target feature recognition method based on artificial intelligence as described in claim 1, characterized in that, The time-frequency transformation of the candidate signal sequence includes: The candidate signal sequence is divided into frames according to a preset window length, and each frame signal is multiplied with a preset weighted window function to obtain a windowed signal. Perform a Fast Fourier Transform (FFT) on the windowed signal to obtain the spectral distribution of each signal; The spectral distributions of each frame are spliced ​​together in chronological order to form a time-frequency matrix, and local instantaneous frequency features are extracted from the time-frequency matrix.

5. The radar target feature recognition method based on artificial intelligence as described in claim 1, characterized in that, The multidimensional feature calculation of the candidate signal sequence includes: The candidate signal sequence is convolved with a preset mother wavelet function to obtain wavelet coefficients at different scales. The wavelet energy distribution of the wavelet coefficients at multiple preset scales is calculated, and the wavelet energy distribution is used to characterize the local characteristics of the signal in the time domain and frequency domain. The wavelet energy distribution is arranged according to the scale factor to construct time-frequency feature maps at multiple preset scales, thereby obtaining the spectral features of the candidate signal sequence.

6. The radar target feature recognition method based on artificial intelligence as described in claim 1, characterized in that, The multidimensional feature calculation of the candidate signal sequence includes: Perform absolute value operation on the candidate signal sequence to obtain the amplitude sequence; Envelope detection is performed on the amplitude sequence to obtain the amplitude envelope; Statistical parameters are calculated based on the amplitude envelope and used as amplitude features of the candidate signal sequence.

7. The radar target feature recognition method based on artificial intelligence as described in claim 1, characterized in that, The multidimensional feature calculation of the candidate signal sequence includes: Perform a Hilbert transform on the candidate signal sequence to obtain the analytic signal; The instantaneous phase is calculated based on the real and imaginary parts of the analyzed signal; The instantaneous phase is expanded to form a continuous phase sequence, which serves as the phase feature of the candidate signal sequence.

8. The radar target feature recognition method based on artificial intelligence as described in claim 1, characterized in that, The multidimensional feature calculation of the candidate signal sequence includes: The Doppler frequency shift is calculated based on the candidate signal sequence at multiple time points to obtain the target velocity; The target acceleration is obtained by performing a first-order difference calculation on the target velocity; Based on the time series of the target velocity and the target acceleration, track stability parameters are calculated as kinematic features of the candidate signal sequence.

9. The radar target feature recognition method based on artificial intelligence as described in claim 1, characterized in that, The step of constructing the multidimensional features into a feature vector according to a preset format includes: The amplitude feature, phase feature, spectral feature, and kinematic feature are normalized to obtain a normalized feature sequence; According to the preset feature mapping relationship, the normalized feature sequence is divided into time domain feature components, frequency domain feature components and spatial domain feature components. The time-domain feature components, the frequency-domain feature components, and the spatial-domain feature components are concatenated in a preset order to form a feature vector of uniform length.

10. The radar target feature recognition method based on artificial intelligence as described in claim 1, characterized in that, The step of inputting the feature vector into a target recognition model pre-trained with a convolutional neural network and a recurrent neural network, and outputting the target feature recognition result, includes: The feature vector is input into the input layer of the CNN, and local spatial features are extracted through convolution and pooling operations. The local spatial features are expanded and input into the RNN, and sequence dependency features are obtained through time series modeling. The sequence-dependent features are input into the fully connected layer and the classification layer, and the corresponding target category label is output to obtain the target feature recognition result.

Images