CN-121995330-A - Radar target feature recognition method based on artificial intelligence

CN121995330ACN 121995330 ACN121995330 ACN 121995330ACN-121995330-A

Abstract

The invention discloses an artificial intelligence-based radar target feature recognition method which comprises the steps of carrying out analog-to-digital conversion on a received echo signal to obtain time domain sampling data of the echo signal, carrying out pretreatment on the time domain sampling data to obtain a candidate signal sequence, carrying out at least one of pulse compression, moving target detection and clutter suppression on the candidate signal sequence, carrying out time-frequency conversion and multi-dimensional feature calculation on the candidate signal sequence, extracting multi-dimensional features of the candidate signal sequence, wherein the multi-dimensional features comprise amplitude features, phase features, frequency spectrum features and kinematic features, constructing the multi-dimensional features into feature vectors according to a preset format, wherein the feature vectors comprise time domain feature components, frequency domain feature components and space domain feature components, inputting the feature vectors into a target recognition model which is trained by a convolutional neural network CNN and a cyclic neural network RNN in advance, and outputting target feature recognition results.

Inventors

ZHANG SHUCHAO
YANG YANG
MAN QINGQING

Assignees

无锡雷视通电子技术有限公司

Dates

Publication Date: 20260508
Application Date: 20250928

Claims (10)

1. The radar target feature recognition method based on artificial intelligence is characterized by comprising the following steps of: performing analog-to-digital conversion on the received echo signal to obtain time domain sampling data of the echo signal; Preprocessing the time domain sampling data to obtain a candidate signal sequence, wherein the preprocessing comprises at least one of pulse compression, moving target detection and clutter suppression; performing time-frequency transformation and multidimensional feature calculation on the candidate signal sequence, and extracting multidimensional features of the candidate signal sequence, wherein the multidimensional features comprise amplitude features, phase features, frequency spectrum features and kinematic features; Constructing the multidimensional feature as a feature vector according to a preset format, wherein the feature vector comprises a time domain feature component, a frequency domain feature component and a space domain feature component; And inputting the feature vector to a target recognition model which is trained by adopting a convolutional neural network CNN and a cyclic neural network RNN in advance, and outputting a target feature recognition result.
2. The artificial intelligence based radar target feature recognition method of claim 1, wherein the preprocessing includes clutter suppression; the preprocessing the time domain sampling data to obtain a candidate signal sequence comprises the following steps: Utilizing a space-time adaptive processing (STAP) algorithm to inhibit fixed clutter in the time domain sampling data to obtain a signal sequence after clutter inhibition; and obtaining a candidate signal sequence based on the signal sequence after clutter suppression.
3. The artificial intelligence-based radar target feature recognition method according to claim 2, wherein the weight vector of the STAP algorithm is iteratively calculated by means of least mean square LMS updating, and the iterative calculation is successively converged based on a covariance matrix of the signal sequence.
4. The artificial intelligence based radar target feature identification method of claim 1, wherein said time-frequency transforming the candidate signal sequence comprises: Framing the candidate signal sequence according to a preset window length, and multiplying each frame of signal by a preset weighting window function to obtain a windowed signal; performing Fast Fourier Transform (FFT) on the windowed signals to obtain frequency spectrum distribution corresponding to each signal; the frequency spectrum distribution of each frame is spliced in time sequence to form a time-frequency matrix, and local instantaneous frequency characteristics are extracted from the time-frequency matrix.
5. The artificial intelligence based radar target feature identification method of claim 1, wherein the performing multidimensional feature computation on the candidate signal sequence comprises: Convolving the candidate signal sequence with a preset mother wavelet function to obtain wavelet coefficients under different scales; Calculating wavelet energy distribution of the wavelet coefficient under a plurality of preset scales, wherein the wavelet energy distribution is used for representing local characteristics of signals in a time domain and a frequency domain; and arranging the wavelet energy distribution according to scale factors, and constructing time-frequency characteristic diagrams under a plurality of preset scales to obtain the frequency spectrum characteristics of the candidate signal sequences.
6. The artificial intelligence based radar target feature identification method of claim 1, wherein the performing multidimensional feature computation on the candidate signal sequence comprises: absolute value operation is carried out on the candidate signal sequence to obtain an amplitude sequence; Performing envelope detection on the amplitude sequence to obtain an amplitude envelope; And calculating a statistical parameter based on the amplitude envelope as an amplitude characteristic of the candidate signal sequence.
7. The artificial intelligence based radar target feature identification method of claim 1, wherein the performing multidimensional feature computation on the candidate signal sequence comprises: Performing Hilbert transformation on the candidate signal sequence to obtain an analysis signal; Calculating an instantaneous phase according to the real part and the imaginary part of the analytic signal; And performing unfolding processing on the instantaneous phase to form a continuous phase sequence serving as the phase characteristic of the candidate signal sequence.
8. The artificial intelligence based radar target feature identification method of claim 1, wherein the performing multidimensional feature computation on the candidate signal sequence comprises: Calculating Doppler frequency shift according to the candidate signal sequences at multiple moments to obtain a target speed; performing first-order differential calculation on the target speed to obtain target acceleration; and calculating a track stability parameter based on the time sequence of the target speed and the target acceleration as the kinematic characteristic of the candidate signal sequence.
9. The method for identifying radar target features based on artificial intelligence according to claim 1, wherein the constructing the multi-dimensional features into feature vectors according to a preset format comprises: normalizing the amplitude characteristic, the phase characteristic, the frequency spectrum characteristic and the kinematic characteristic to obtain a normalized characteristic sequence; Dividing the normalized feature sequence into a time domain feature component, a frequency domain feature component and a space domain feature component according to a preset feature mapping relation; And splicing the time domain feature components, the frequency domain feature components and the space domain feature components according to a preset arrangement sequence to form feature vectors with uniform lengths.
10. The method for identifying radar target features based on artificial intelligence according to claim 1, wherein inputting the feature vector into a target identification model trained in advance by using a convolutional neural network and a cyclic neural network, and outputting a target feature identification result, comprises: inputting the feature vector to an input layer of the CNN, and extracting local space features through convolution operation and pooling operation; The local spatial features are input to the RNN after being unfolded, and sequence dependent features are obtained through time sequence modeling; And inputting the sequence dependent features to a full-connection layer and a classification layer, and outputting corresponding target class labels to obtain target feature recognition results.

Description

Radar target feature recognition method based on artificial intelligence Technical Field The invention relates to the technical field of photoelectricity, in particular to a radar target feature recognition method based on artificial intelligence. Background With the development of electronic countermeasure and radar detection technologies, radar target feature identification has important significance in military and civil scenarios. The traditional direction finding and identifying method mainly comprises multi-beam amplitude-comparison direction finding and interferometer direction finding. The multi-beam amplitude-comparison direction finding determines the arrival angle of a target by comparing the relative amplitudes of the received signals of adjacent beams, but has strict requirements on beam design and is easy to cause nonlinear errors by side lobes. The interferometer direction finding calculates the arrival angle through the phase difference of two paths of received signals, has a simple structure, and is easily limited by phase blurring and image blurring. Although these methods are improved, they are limited to azimuth measurement, and it is difficult to realize spectrum feature identification at the same time, and there are disadvantages in complex electromagnetic environments. Chinese patent No. CN114325628a discloses a method and apparatus for identifying radar target features based on optical delay fast interference scanning, which uses an electro-optical converter and an optical delay digital control scanning apparatus to perform delay control on a received signal, implements signal synthesis through optical combination and photoelectric detection, and calculates radio frequency signal power by a power feature analysis apparatus. Through forward and reverse light delay scanning, the corresponding relation between the power peak value and the delay amount is analyzed, the measurement of the azimuth angle, the frequency and the frequency spectrum characteristics of the target signal is realized, and the method is suitable for radar target detection under different electromagnetic environments and modulation systems. However, the above prior art still relies on optical delay hardware and power detection to realize target recognition, and has the problems that signal processing relies on a physical device, a feature extraction mode is single, and the method is difficult to adapt to changeable complex signal environments. Disclosure of Invention Aiming at the technical problems that the prior art relies on optical delay hardware and power detection to realize target identification, the signal processing relies on a physical device, the characteristic extraction mode is single, and the method is difficult to adapt to changeable complex signal environments, the invention provides an artificial intelligence-based radar target characteristic identification method. In order to solve the technical problems, the invention provides the following technical scheme: The invention provides a radar target feature recognition method based on artificial intelligence, which comprises the following steps: performing analog-to-digital conversion on the received echo signal to obtain time domain sampling data of the echo signal; Preprocessing the time domain sampling data to obtain a candidate signal sequence, wherein the preprocessing comprises at least one of pulse compression, moving target detection and clutter suppression; performing time-frequency transformation and multidimensional feature calculation on the candidate signal sequence, and extracting multidimensional features of the candidate signal sequence, wherein the multidimensional features comprise amplitude features, phase features, frequency spectrum features and kinematic features; Constructing the multidimensional feature as a feature vector according to a preset format, wherein the feature vector comprises a time domain feature component, a frequency domain feature component and a space domain feature component; And inputting the feature vector to a target recognition model which is trained by adopting a convolutional neural network CNN and a cyclic neural network RNN in advance, and outputting a target feature recognition result. In an alternative embodiment, the preprocessing includes clutter suppression; the preprocessing the time domain sampling data to obtain a candidate signal sequence comprises the following steps: Utilizing a space-time adaptive processing (STAP) algorithm to inhibit fixed clutter in the time domain sampling data to obtain a signal sequence after clutter inhibition; and obtaining a candidate signal sequence based on the signal sequence after clutter suppression. In an alternative embodiment, the weight vector of the STAP algorithm is iteratively calculated by a least mean square LMS update mode, and the iterative calculation is successively converged based on a covariance matrix of the signal sequence. In an alternative