A target recognition method based on time-reversal one-dimensional range profile
Through time inversion technology and feature extraction, the problem of large amount of data and small difference between strength and weakness scattering of traditional one-dimensional high-resolution distance images is solved, and the efficiency and accuracy of radar target recognition are improved, achieving more efficient recognition effects.
Patent Information
- Application Number
- CN202210117101.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-02-07
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2042-02-07
AI Technical Summary
Traditional one-dimensional high-resolution distance images have large data volumes in broadband radars and there is little difference between strong and weak scattering, resulting in low target recognition efficiency and accuracy.
Time inversion technology is used to obtain the time inversion one-dimensional distance image, and the central moment feature and waveform entropy feature are extracted through inverse Fourier transform and normalization processing, and the recognition is combined with a support vector machine.
The data calculation amount is reduced, the recognition efficiency and accuracy are improved, the recognition calculation amount and storage difficulty are reduced, and the feature separation of strong and weak scattering points is enhanced.
Smart Images

Figure CN114609600B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of signal processing, and in particular relates to a target recognition method based on time-inverted one-dimensional range image. Background Art
[0002] Radar target recognition is an important development area of modern radar technology, and plays a vital role in improving radar information extraction and analysis capabilities. The principle of radar target recognition is to judge and extract the characteristic information of the target by analyzing all echoes, and to judge the attribute category of the target by different prior information of different targets.
[0003] In the prior art, high-altitude target recognition is usually performed by extracting features using a traditional one-dimensional high-resolution range profile (HRRP). A one-dimensional high-resolution range profile can reflect the structure and shape of the target and display the distribution of the target's scattering points along the distance direction, providing important information for target recognition.
[0004] However, there are still some problems with the traditional one-dimensional high-resolution range profile for target recognition. First, in wideband radar, in order to obtain higher resolution, ultra-wideband is usually used, which will increase the amount of HRRP data; second, in some cases, the difference between strong scattering and weak scattering is not much, which will make recognition more difficult and result in larger recognition errors. Summary of the invention
[0005] In order to solve the above problems existing in the prior art, the present invention provides a target recognition method based on time-reversed one-dimensional range image. The technical problem to be solved by the present invention is achieved by the following technical solutions:
[0006] A target recognition method based on time-reversed one-dimensional range image, the method comprising: step 1: constructing a data set based on multiple high-altitude targets; wherein the data set comprises: a training data set and a test data set; step 2: based on the data set, obtaining the primary echo data corresponding to each high-altitude target; wherein the primary echo data comprises: primary echo training data and primary echo test data; step 3: calculating the primary echo data by time-reversal technology to obtain the time-reversed echo data corresponding to each high-altitude target; wherein the time-reversed echo data comprises: time-reversed echo training data and time-reversed echo test data; step 4: performing inverse Fourier transform and taking absolute value on the time-reversed echo data respectively to obtain time-reversed one-dimensional range image data; wherein the time-reversed one-dimensional range image data is obtained. The image data includes: time-reversed one-dimensional range image training data and time-reversed one-dimensional range image test data; step 5: normalizing the time-reversed one-dimensional range image data, and extracting central moment features and waveform entropy features based on the normalized time-reversed one-dimensional range image data; step 6: splicing the central moment features and waveform entropy features corresponding to the time-reversed one-dimensional range image training data to obtain a first fusion feature; splicing the central moment features and waveform entropy features corresponding to the time-reversed one-dimensional range image test data to obtain a second fusion feature; step 7: based on a support vector machine, training the first fusion features corresponding to all high-altitude targets to obtain a training network; step 8: respectively identifying the second fusion features corresponding to each high-altitude target with the training network to obtain the identification result corresponding to each high-altitude target.
[0007] In one embodiment of the present invention, the data set includes a plurality of transmission signal spectra S(ω q ), the target frequency response H(ω q ), additive white Gaussian noise V(ω q ), where q = 0, 1, ..., Q-1; the step 2 comprises: based on the data set, obtaining the primary echo data Y(ω corresponding to different high-altitude targets q ), expressed as:
[0008] Y(ω q )=S(ω q )H(ω q )+V(ω q ).
[0009] In one embodiment of the present invention, step 3 includes: step 3-1: using time inversion technology, in the frequency domain, the primary echo data Y(ω q ) is conjugated and expressed as: * (ω q ); Step 3-2: For Y* (ω q ) is subjected to energy standardization and is expressed as: kY * (ω q ), k is the energy normalization coefficient;
[0010]
[0011] Step 3-3: kY * (ω q ) is launched, and the frequency response H(ω q ) is used to obtain the echo data X(ω q ), expressed as: X(ω q )=kY * (ω q )H(ω q )+W(ω q ), where W(ω q ) is additive white Gaussian noise.
[0012] Beneficial effects of the present invention:
[0013] On the one hand, compared with the conventional one-dimensional range profile (HRRP) recognition method in the prior art, when the data volume of the one-dimensional range profile is large, or the difference between strong scattering and weak scattering is not large, the recognition efficiency and accuracy of the target are low. The present invention introduces the time reversal technology, and can utilize the symmetry of the time reversal one-dimensional range profile obtained by the time reversal technology. In the process of target recognition, only half of the data needs to be selected, thereby greatly reducing the amount of data calculation and improving the recognition efficiency; and, compared with the conventional one-dimensional range profile, the time reversal one-dimensional range profile has the characteristics of stronger strong scattering points, weaker weak scattering points, separate strong and weak features, and great difference between strong and weak scattering points, which can improve the accuracy of target recognition.
[0014] On the other hand, after the present invention uses the time reversal technology to obtain the time-reversed one-dimensional range image, in order to avoid the natural time-shift sensitivity of the one-dimensional range image, the central moment feature and waveform entropy feature of the time-reversed one-dimensional range image are extracted, and the fusion feature is obtained by splicing to serve as the feature of target recognition. The central moment feature and the waveform entropy feature have the shape information feature and energy distribution feature of the one-dimensional range image respectively, and the data volume is much smaller than the data volume of the one-dimensional range image, so that the present invention can greatly reduce the target recognition calculation amount and storage difficulty, so as to improve the accuracy and efficiency of target recognition.
[0015] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1Schematic diagram of a target recognition method based on time-reversed one-dimensional range profile provided by an embodiment of the present invention;
[0017] Figure 2 It is a schematic diagram of the comparison of the amplitude of a one-dimensional range image and a time-reversed one-dimensional range image of a target recognition method based on a time-reversed one-dimensional range image provided by an embodiment of the present invention;
[0018] Figure 3 It is a schematic diagram of the relationship between the recognition rate and the signal-to-noise ratio of a target recognition method based on one-dimensional range profile provided by an embodiment of the present invention.
[0019] Figure 4 This is a relationship diagram of the recognition rate versus signal-to-noise ratio of a target recognition method based on time-reversed one-dimensional range profile provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0020] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.
[0021] Example
[0022] See also Figure 1 , Figure 1 : is a schematic diagram of a target recognition method based on time-reversed one-dimensional range profile provided by an embodiment of the present invention, the method comprising:
[0023] Step 1: construct a data set based on multiple high-altitude targets; wherein the data set includes: a training data set and a test data set.
[0024] Optionally, the data set includes a plurality of emission signal spectra S(ω q ), the target frequency response H(ω q ), additive white Gaussian noise V(ω q ), where q=0,1,…,Q-1.
[0025] For example, the present invention uses the data of different high-altitude targets F22, B2, and MQ9 obtained by Feko software to divide them into a training data set and a test data set, and ensures that the two types of data sets are not repeated.
[0026] Step 2: Based on the data set, obtain the primary echo data corresponding to each high-altitude target; wherein the primary echo data includes: primary echo training data and primary echo test data.
[0027] Optionally, the step 2 includes:
[0028] Based on the data set, obtain the primary echo data Y(ω) corresponding to different high-altitude targets q ), expressed as:
[0029] Y(ω q )=S(ω q )H(ω q )+V(ω q )
[0030] For example, the present invention uses Feko software to construct the target frequency response H(ω) of three types of high-altitude targets F22, B2, and MQ9 q ), transmit Gaussian pulse signal S(ω q ), S(ω q ) and the target frequency response H(ω q ) is used, and the first echo Y(ω q )=S(ω q )H(ω q )+V(ω q ).
[0031] Step 3: Calculate the primary echo data through time reversal technology to obtain the echo data after time reversal corresponding to each high-altitude target; wherein the echo data after time reversal includes: echo training data after time reversal and echo test data after time reversal.
[0032] The present invention combines the time inversion technology for realizing adaptive spatial and temporal focusing with target recognition. The time inversion echo is symmetrical in the time domain, and only half of the time inversion echo needs to be intercepted for target recognition, thereby greatly reducing the amount of calculation and storage. Time inversion can also enhance the amplitude of strong scattering and reduce the amplitude of weak scattering, making the features more obvious and facilitating subsequent recognition. Feature extraction is an important component of the radar target recognition system in the target recognition stage. Furthermore, the present invention extracts the central moment feature and waveform entropy feature from the target HRRP (high resolution range profile). The central moment feature is a time-shift invariant feature that can reflect the shape information of the target. It is a simple and effective radar echo waveform feature that can be directly used for radar target recognition; the waveform entropy feature can represent the energy distribution in the target signal and can represent the uniformity of the radial distribution of the signal energy.
[0033] Optionally, step 3 includes:
[0034] Step 3-1: Use time reversal technology to perform a single echo data Y(ω q ) is conjugated and expressed as: * (ω q ).
[0035] Step 3-2: For Y * (ωq ) is subjected to energy standardization and is expressed as: kY * (ω q ), k is the energy normalization coefficient;
[0036]
[0037] Step 3-3: kY * (ω q ) is launched, and the frequency response H(ω q ) is used to obtain the echo data X(ω q ), expressed as:
[0038] X(ω q )=kY * (ω q )H(ω q )+W(ω q ),
[0039] Among them, W(ω q ) is additive white Gaussian noise.
[0040] Step 4: Perform inverse Fourier transform and take absolute value on the echo data after time inversion to obtain time inverted one-dimensional range image data; wherein the time inverted one-dimensional range image data includes: time inverted one-dimensional range image training data and time inverted one-dimensional range image test data.
[0041] Step 5: Normalize the time-reversed one-dimensional range image data, and extract the central moment feature and waveform entropy feature based on the normalized time-reversed one-dimensional range image data.
[0042] Optionally, the time-reversed one-dimensional range image data is represented as |x(t)|, wherein x(t) represents the time domain signal of the time-reversed echo.
[0043] Optionally, step 5 includes:
[0044] Step 5-1: Normalize the time-reversed one-dimensional range image data |x(t)|, expressed as:
[0045]
[0046] Where t = 1, 2, ..., N, N is the signal dimension;
[0047] Step 5-2: Extraction The central moment characteristic C t , expressed as:
[0048]
[0049] in, is the first-order origin moment, indicating the scattering center, and p indicates the order;
[0050] Step 5-3: Extraction The waveform entropy feature H t , expressed as:
[0051]
[0052] Step 6: Concatenate the central moment feature and waveform entropy feature corresponding to the time-reversed one-dimensional range image training data to obtain the first fusion feature; concatenate the central moment feature and waveform entropy feature corresponding to the time-reversed one-dimensional range image test data to obtain the second fusion feature.
[0053] Optionally, the first fusion feature is represented as e t =[C t ,H t ], the second fusion feature is represented as e t =[C t ,H t ].
[0054] Step 7: Based on the support vector machine, the first fusion features corresponding to all high-altitude targets are trained to obtain a training network.
[0055] Optionally, denote the trained network as M.
[0056] For example, the support vector machine is a Gaussian kernel support vector machine. The present invention can import the first fusion features corresponding to all high-altitude targets into the support vector machine for training to obtain a training network.
[0057] Step 8: The second fusion feature corresponding to each high-altitude target is respectively processed with the training network to obtain the recognition result corresponding to each high-altitude target.
[0058] It should be noted that the present invention may also be: after obtaining the first fusion feature e corresponding to the training data set t1 Afterwards, obtain the second fusion feature e corresponding to the test data set t2 , then based on the first fusion feature e t1 Obtain the training network, and finally perform recognition processing based on the training network to obtain the recognition results corresponding to each high-altitude target.
[0059] In summary, the beneficial effects of the present invention are:
[0060] On the one hand, compared with the conventional one-dimensional range profile (HRRP) recognition method in the prior art, when the data volume of the one-dimensional range profile is large, or the difference between strong scattering and weak scattering is not large, the recognition efficiency and accuracy of the target are low. The present invention introduces the time reversal technology, and can utilize the symmetry of the time reversal one-dimensional range profile obtained by the time reversal technology. In the process of target recognition, only half of the data needs to be selected, thereby greatly reducing the amount of data calculation and improving the recognition efficiency; and, compared with the conventional one-dimensional range profile, the time reversal one-dimensional range profile has the characteristics of stronger strong scattering points, weaker weak scattering points, separate strong and weak features, and great difference between strong and weak scattering points, which can improve the accuracy of target recognition.
[0061] On the other hand, after the present invention uses the time reversal technology to obtain the time-reversed one-dimensional range image, in order to avoid the natural time-shift sensitivity of the one-dimensional range image, the central moment feature and waveform entropy feature of the time-reversed one-dimensional range image are extracted, and the fusion feature is obtained by splicing to serve as the feature of target recognition. The central moment feature and the waveform entropy feature have the shape information feature and energy distribution feature of the one-dimensional range image respectively, and the data volume is much smaller than the data volume of the one-dimensional range image, so that the present invention can greatly reduce the target recognition calculation amount and storage difficulty, so as to improve the accuracy and efficiency of target recognition.
[0062] Furthermore, combined with simulation experiments, the beneficial effects of the present invention are verified:
[0063] 1) Simulation conditions: The transmitted signal is a Gaussian pulse with a center frequency of 17.5 GHz and a signal bandwidth of 1 GHz. The simulation targets are three high-altitude target aircraft. The experimental data processing is completed on MATLAB 2019, and the channel response of the target is simulated on FEKO.
[0064] 2) Simulation content:
[0065] Simulation 1: Using the target frequency response obtained by FEKO, a Gaussian pulse is emitted to obtain a one-dimensional range image, and then the time reversal technique is used to obtain a time-reversed one-dimensional range image. The amplitude comparison between the one-dimensional range image and the time-reversed one-dimensional range image is as follows: Figure 2 shown.
[0066] Simulation 2: Extract features for recognition experiments on one-dimensional range images. The signal-to-noise ratio of the training data is fixed at 30dB, and the signal-to-noise ratio of the test data is 10-60dB. The average recognition rate and the recognition rate of each of the three types of targets vary with the signal-to-noise ratio. Figure 3 shown.
[0067] Simulation 3: Extract features from the time-reversed one-dimensional range image and conduct recognition experiments. The signal-to-noise ratio of the training data is fixed at 30dB, and the signal-to-noise ratio of the test data is 10-60dB. The average recognition rate and the recognition rate of each of the three types of targets vary with the signal-to-noise ratio. Figure 4 shown.
[0068] 3) Analysis of simulation results:
[0069] Under the same signal-to-noise ratio, the target recognition rate of the target recognition experiment based on the time-inverted one-dimensional range image proposed by the present invention is significantly better than that of the one-dimensional range image.
[0070] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.
Claims
1. A target recognition method based on time-reversed one-dimensional range profile, characterized in that: The method comprises: Step 1: construct a data set based on multiple high-altitude targets; wherein the data set includes: a training data set and a test data set; Step 2: Based on the data set, obtain the primary echo data corresponding to each high-altitude target; wherein the primary echo data includes: primary echo training data and primary echo test data; Step 3: Calculate the primary echo data by time reversal technology to obtain the echo data after time reversal corresponding to each high-altitude target; wherein the echo data after time reversal includes: echo training data after time reversal and echo test data after time reversal; Step 4: Perform inverse Fourier transform and take absolute value on the echo data after time inversion to obtain time inversion one-dimensional range image data; wherein the time inversion one-dimensional range image data includes: time inversion one-dimensional range image training data and time inversion one-dimensional range image test data; Step 5: normalize the time-reversed one-dimensional range image data, and extract the central moment feature and waveform entropy feature based on the normalized time-reversed one-dimensional range image data; Step 6: splicing the central moment feature and waveform entropy feature corresponding to the time-reversed one-dimensional range image training data to obtain the first fusion feature; splicing the central moment feature and waveform entropy feature corresponding to the time-reversed one-dimensional range image test data to obtain the second fusion feature; Step 7: Based on the support vector machine, the first fusion features corresponding to all high-altitude targets are trained to obtain a training network; Step 8: Perform recognition processing on the second fusion features corresponding to each high-altitude target with the training network respectively to obtain the recognition result corresponding to each high-altitude target; or, after obtaining the first fusion features corresponding to the training data set, obtain the second fusion features corresponding to the test data set, and then obtain the training network based on the first fusion features, and finally perform recognition processing based on the training network to obtain the recognition result corresponding to each high-altitude target.
2. The method according to claim 1, characterized in that The data set includes the emission signal spectra S(ω) corresponding to multiple high-altitude targets q ), the target frequency response H(ω q ), additive white Gaussian noise V(ω q ), where q = 0, 1, ..., Q-1; The step 2 comprises: Based on the data set, obtain the primary echo data Y(ω) corresponding to different high-altitude targets q ), expressed as: Y(ω q )=S(ω q )H(ω q )+V(ω q )。 3. The method according to claim 2, characterized in that The step 3 comprises: Step 3-1: Use time reversal technology to perform a single echo data Y(ω q ) is conjugated and expressed as: * (ω q ); Step 3-2: For Y * (ω q ) is subjected to energy standardization and is expressed as: kY * (ω q ), k is the energy normalization coefficient; Step 3-3: kY * (ω q ) is launched, and the frequency response H(ω q ) is used to obtain the echo data X(ω q ), expressed as: X(ω q )=kY * (oh q )H(ω q )+W(ω q ), Among them, W(ω q ) is additive white Gaussian noise.
4. The method according to claim 3, characterized in that The time-reversed one-dimensional range image data is represented as |x(t)|, wherein x(t) represents the time domain signal of the time-reversed echo; the step 5 comprises: Step 5-1: Normalize the time-reversed one-dimensional range image data |x(t)|, expressed as: Where t = 1, 2, ..., N, N is the signal dimension; Step 5-2: Extraction The central moment characteristic C t , expressed as: in, is the first-order origin moment, indicating the scattering center, and p indicates the order; Step 5-3: Extraction The waveform entropy feature H t , expressed as:
5. The method according to claim 4, characterized in that The first fusion feature is expressed as The second fusion feature is expressed as
Citation Information
Patent Citations
radar HRRP target recognition method based on priori optimal variational autoencoder
CN110109110A
High-resolution range profile target identification method based on signal-to-noise ratio domain knowledge network
CN112882010A
