Micromotion extraction and scattering point identification method of smooth precessing cone based on narrow band

Abstract

The present invention discloses a method for extracting micro-motion and identifying scattering points of a smooth precession cone based on narrowband. Based on narrowband radar, an equivalent scattering point model of a smooth precession cone target is established, the number of echo signals is estimated, the sliding window Pole Root-MUSIC algorithm is ordered, the window length is set, the sliding window estimates the micro-Doppler time-frequency curve of the target scattering point and performs IRadon transformation, the energy concentration is located and identified, the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point are distinguished, and the scattering point identification is realized; then Radon transformation is performed respectively to obtain two corresponding trend lines, and the micro-Doppler time-frequency curves of the target scattering point are respectively associated and smoothed according to the trend lines, so as to realize the micro-motion feature extraction of the target. Under the condition of narrowband radar observation, the present invention can directly and quickly obtain the micro-Doppler time-frequency curve and complete the association.

Description

Technical Field

[0001] The invention belongs to the technical field of micro-Doppler signal extraction, and in particular relates to a method for extracting micro-motion and identifying scattering points of a narrow-band smooth precession cone. Background Art

[0002] The smooth precessing cone is a typical radar target, and its micromotion mainly includes spin, precession, nutation, swing, tumble, etc. These micromotions produce periodically modulated micro-Doppler effect under radar observation. Therefore, micromotion feature extraction and micro-Doppler analysis are important links in the recognition of smooth precessing cone targets.

[0003] At present, the echo analysis of spatial cone targets is mostly based on time-frequency analysis. Since the target micro-motion is periodic, the radar echo and time-frequency curve are both periodic, and the micro-motion characteristics of the target can be extracted from the time-frequency curve data. However, the existing time-frequency analysis methods, such as short-time Fourier transform and wavelet transform, can only obtain low-resolution time-frequency graphs, and related algorithms are still needed to extract time-frequency curves. At the same time, the existing time-frequency curve extraction methods have the problems of difficulty in extracting multi-target curves and irrelevance. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a method for micro-motion extraction and scattering point identification based on a narrow-band smooth precession cone, which can directly and quickly obtain time-frequency curves and complete association under narrow-band radar observation.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for extracting micro-motion and identifying scattering points of a narrow-band smooth precession cone comprises the following steps:

[0007] Step 1: Based on narrow-band radar, an equivalent scattering point model of a smooth precessing cone target is established;

[0008] Step 2, based on the equivalent scattering point model, the number of echo signals is estimated, the sliding window Pole Root-MUSIC algorithm is ordered, the window length of the sliding window Pole Root-MUSIC algorithm is set, and the micro-Doppler time-frequency curve of the target scattering point is estimated by sliding the window; the target scattering point is two scattering points of the smooth precession cone target under radar observation;

[0009] Step 3, perform IRadon transformation on the micro-Doppler time-frequency curve of the target scattering point obtained by sliding window estimation, locate and identify the energy concentration of the IRadon transformation, distinguish the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point, and realize scattering point identification;

[0010] Step 4, performing Radon transformation on the micro-Doppler time-frequency curves of the distinguished cone top scattering point and the cone bottom scattering point to obtain two corresponding trend lines;

[0011] Step 5: Correlate and smooth the micro-Doppler time-frequency curves of the target scattering points according to the two trend lines to extract the micro-motion features of the target.

[0012] In one embodiment, the step 2 is to set the window length of the sliding window Pole Root-MUSIC algorithm and use the sliding window to estimate the micro-Doppler time-frequency curve of the target scattering point, and the implementation method is as follows:

[0013] dividing the echo signal into overlapping signal sub-segments;

[0014] The signal sub-fragments are discretized into points, and obtain a discrete signal and its covariance matrix, which includes the true matrix and the estimation matrix ;

[0015] According to the real matrix and the estimation matrix , calculate the discrete signal The estimated instantaneous micro-Doppler frequency , and according to A micro-Doppler time-frequency curve characterizing the target scattering point.

[0016] In one embodiment, the echo signal is divided into overlapping signal sub-segments, wherein the first Signal subfragment It is expressed as:

[0017]

[0018] In the formula, It is The amplitude of the signal sub-segment, K is the number of signal sub-segments, It is The instantaneous micro-Doppler frequency of the signal sub-segment, For slow time, It is The instantaneous phase of a signal sub-segment, is the additive white noise of the smooth precession cone target echo signal, j is the imaginary number sign;

[0019] The discrete signal , expressed as:

[0020]

[0021] In the formula, The discretization of the signal sub-segment points;

[0022] The real matrix , expressed as:

[0023]

[0024] The estimation matrix , expressed as:

[0025]

[0026] In the formula, is a The matrix consists of the eigenvectors corresponding to the signal subspace, express The conjugate transposed matrix of is a Matrix, represented by The diagonal matrix composed of the eigenvalues ​​of represents the noise power, express Identity matrix;

[0027] represents the signal subspace, represents the noise subspace, represents the eigenvalue belonging to the signal subspace, represents the eigenvalue belonging to the noise subspace, express The conjugate transposed matrix of express The conjugate transposed matrix of ;

[0028] , , represents the steering vector, , yes of Power, For polynomial roots on the unit circle;

[0029] The discrete signal The estimated instantaneous micro-Doppler frequency , expressed as:

[0030]

[0031] In the formula, Represents the solution of the equation The roots obtained, express phase.

[0032] In one embodiment, in step 3, the micro-Doppler time-frequency curve of the target scattering point estimated by the sliding window is denoised and mapped into a two-dimensional image, and the two-dimensional image is subjected to IRadon transformation;

[0033] The denoising is implemented as follows:

[0034] The noise is located by using the histogram method and the differential method respectively. If the noise located by the histogram method is not located by the differential method, the noise is false noise; if the noise located by the histogram method is also located by the differential method, the noise is real noise, and the real noise is denoised.

[0035] In one embodiment, the two-dimensional image is subjected to IRadon transformation, and the angle The IRadon domain is represented as:

[0036]

[0037] in, represents a two-dimensional image, represents a sine curve, , is the initial coordinate determined by the initial stage of the micro-Doppler time-frequency curve, is the target precession angular frequency of the smooth precession cone, For slow time, , represents the total observation time, represents the filter, is the integration variable.

[0038] In one embodiment, the step 3 is to locate and identify the IRadon transformation energy concentration and distinguish the micro-Doppler time-frequency curves of the cone top and the cone bottom, and the implementation method is as follows:

[0039] For different initial phases A recognition model is trained, wherein the scattering points at the top of the cone correspond to the area where energy is concentrated; the scattering points at the bottom of the cone correspond to the area where energy is dispersed, and are symmetrically distributed about the line connecting the top of the cone and the center of the two-dimensional image; the recognition model obtained by training is used to realize the positioning and recognition of the image domain signals of the top and bottom of the cone after IRadon transformation.

[0040] In one embodiment, in step 4, Radon transform is performed on the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point respectively to obtain two corresponding trend lines, and the method is as follows:

[0041] For the micro-Doppler time-frequency curve of the cone top scattering point, a trend line in the form of a sine curve is obtained by Radon transformation. ;

[0042] For the micro-Doppler time-frequency curve of the cone bottom scattering point, Radon transform is performed to obtain two trend lines in the form of sine curves. and ,Will and The weighted summation obtains a trend line that is closest to the micro-Doppler time-frequency curve of the cone bottom scattering point. .

[0043] In one embodiment, for positioning the position , the trend line it represents is obtained through Radon transformation:

[0044]

[0045] in is the initial phase of the trend line, which is expressed as follows:

[0046]

[0047] is the trend line amplitude, which is expressed as follows:

[0048] represents the location coordinates of the energy focus in the image domain after IRadon transformation, which is set the center of

[0049] when When the value detection is located in the cone top image domain, ,when When the value detection is located in the cone bottom image domain, .

[0050] In one embodiment, the two trend lines and have the same amplitude and frequency, and the weight is 0.5. and Weighted summation is performed to obtain the trend line that is closest to the amplitude and frequency of the micro-Doppler time-frequency curve of the cone bottom scattering point. .

[0051] In one embodiment, the micro-Doppler time-frequency curves of the target scattering points are associated and smoothed by a nearest neighbor algorithm.

[0052] Most of the existing time-frequency analysis methods are non-parametric methods with limited resolution, which is not conducive to further application. In contrast, the present invention proposes a method for extracting micro-Doppler time-frequency information and identifying scattering point types for smoothly precessing cone targets under narrowband radar observation. For the echo of a smoothly precessing cone target composed of multi-component signals, the sliding window Pole Root-MUSIC algorithm is used to extract the instantaneous micro-Doppler frequency of the target, and the IRadon transform and recognition model are used to locate, identify and associate the extracted instantaneous micro-Doppler frequency, completing the reconstruction of the micro-Doppler frequency of the target scattering point, and realizing the distinction and identification of the cone top and cone bottom signals. The estimation accuracy is not limited by the sampling interval and is much higher than that of the non-parametric method. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the equivalent scattering point model of a smoothly precessing cone target.

[0054] Figure 2 Schematic diagram of the simulation results of the micro-motion narrow-band radar echo signal of a smoothly precessing cone target in the equivalent scattering point model.

[0055] Figure 3 This is a schematic diagram of the impact of signal number determination errors on micro-Doppler information extraction, where the horizontal axis is slow time and the vertical axis is frequency.

[0056] Figure 4 This is a schematic diagram of the impact of correctly determining the number of signals on the extraction of micro-Doppler information, where the horizontal axis is slow time and the vertical axis is frequency.

[0057] Figure 5 This is a schematic diagram of the instantaneous time-frequency analysis results of the sliding window Pole Root-MUSCI algorithm when the signal-to-noise ratio is low. The horizontal axis in the figure is slow time and the vertical axis is frequency.

[0058] Figure 6 This is a schematic diagram of the instantaneous time-frequency analysis results of the sliding window Pole Root-MUSCI algorithm when the signal-to-noise ratio is high. The horizontal axis in the figure is slow time and the vertical axis is frequency.

[0059] Figure 7 It is a schematic diagram of the time-frequency analysis results of a signal with a signal-to-noise ratio of 0 dB, specifically showing the STFT time-frequency curve and the instantaneous micro-Doppler time-frequency curve extracted based on the sliding window Pole Root-MUSIC algorithm.

[0060] Figure 8 This is a schematic diagram of the time-frequency analysis results of a signal with a signal-to-noise ratio of 0dB, specifically showing Figure 7 Outlier spurious points at the intersection of the mid-signal.

[0061] Fig. 9This is a schematic diagram of the principle of locating the noise position in the time-frequency curve, from top to bottom: histogram method, difference method and final noise position.

[0062] Fig.10 Schematic diagram of denoising processing using the histogram method to locate noise in the lower part of the time-frequency curve.

[0063] Fig.11 Schematic diagram of denoising processing for noise localization using the difference method in the lower half of the time-frequency curve.

[0064] Fig.12 It is a schematic diagram of the lower half of the time-frequency curve after denoising using the method of the present invention.

[0065] Fig.13 Schematic diagram of denoising processing using the histogram method to locate noise in the upper part of the time-frequency curve.

[0066] Fig.14 Schematic diagram of denoising processing using the difference method for noise localization in the upper part of the time-frequency curve.

[0067] Fig.15 It is a schematic diagram of the upper part of the time-frequency curve after denoising using the method of the present invention.

[0068] Fig.16 It is the model micro-Doppler information extracted based on the sliding window Pole Root-MUSIC algorithm.

[0069] Fig.17 for Fig.16 The IRadon transform domain of the extracted model micro-Doppler information, where regions 1 and 2 are shown.

[0070] Fig.18 for Fig.17 A locally enlarged schematic diagram of the IRadon distribution at the top of the cone in the middle area 1.

[0071] Fig.19 for Fig.17 A locally enlarged schematic diagram of the IRadon distribution at the bottom of cone 2 in the middle area.

[0072] Fig. 20 Schematic diagram of YOLO detection and labeling of IRadon transform.

[0073] Fig.21 For Fig. 20 Schematic diagram of the time-frequency curve obtained by performing Radon transform on the marked results.

[0074] Fig. 22 It is a schematic diagram of the time-frequency curve associated with the time-frequency trend line.

[0075] Fig.23 for Fig. 22Schematic diagram of the comparison between the associated time-frequency curve and the STFT result. DETAILED DESCRIPTION

[0076] The embodiments of the present invention are described in detail below with reference to the accompanying drawings and examples.

[0077] The present invention is a method for extracting micro-Doppler and identifying scattering points of a smooth precessing cone under narrowband radar observation. The algorithm uses sliding window Pole Root-MUSIC (Pole Root-MUltiple SIgnal Classification) to extract instantaneous micro-Doppler information from narrowband radar echo signals, uses IRadon transform and recognition model detection to identify the micro-Doppler distribution of the cone top and cone bottom, and further uses Radon transform to reconstruct the micro-Doppler time-frequency information of the target scattering point.

[0078] The present invention mainly comprises the following specific steps:

[0079] Step 1: Based on the narrowband radar, an equivalent scattering point model of a smoothly precessing cone target is established to derive the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point. The purpose of this step is to verify the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point finally extracted.

[0080] Narrowband radar, with its frequency resolution, simplified system requirements and applicability to motion characteristics extraction, is a suitable choice for analyzing the equivalent scattering point model and micro-Doppler characteristics of a smoothly precessing cone target. The equivalent scattering point model of a smoothly precessing cone target established by the present invention is as follows: Figure 1 As shown, the target height of the smooth precession cone is , the radius of the base of the smooth precession cone target is , Point is the target mass center of the smooth precession cone, and its distance from the bottom of the cone is , the half cone angle is , is the radar line of sight direction, is the angle between the radar line of sight and the target center axis, is the precession angle, and the angle between the radar line of sight and the precession axis is When the space cone target is flying in the middle stage, the radar is generally irradiated head-on. According to the shielding effect, only Figure 1 The cone top scattering point in Scattering point with cone bottom It can be seen that the scattering point at the bottom of the cone is the intersection of the plane formed by the radar line of sight and the target symmetry axis and the edge of the target bottom surface. Assume that the radar line of sight is plane, the radar line of sight is expressed as

[0081]

[0082] Target around Shaft precession, precession angular frequency , spins around the symmetry axis, and the spin angular frequency is , the initial moment is at Axis direction, the target symmetry axis is expressed as:

[0083]

[0084] The angle between the radar line of sight and the symmetry axis of the smoothly precessing cone target is expressed as:

[0085]

[0086] make , , and are the intermediate parameters used in the simplified equation, then

[0087]

[0088] In radar systems, the equivalent scattering point model is used to describe the spatial structure and micro-motion characteristics of the target. The radial distance from the equivalent scattering point on the precessing cone target to the radar is expressed as:

[0089]

[0090]

[0091]

[0092] , , are the projection distances of the three scattering points A, B, and C of the smooth precession cone target on the radar line of sight, is the initial distance between the smooth precession cone target coordinate system and the radar.

[0093] Points are usually not observable due to occlusion.

[0094] Cone top scattering point The micro-Doppler time-frequency curve is composed of its micro-Doppler frequency The formula is as follows:

[0095]

[0096] Cone bottom scattering point The micro-Doppler time-frequency curve is composed of its micro-Doppler frequency The formula is as follows:

[0097]

[0098] From formula (8), we can see that the micro-Doppler time-frequency curve of the cone top scattering point is a sine curve, and from formula (9), we can see that the micro-Doppler time-frequency curve of the cone bottom scattering point is a quasi-sine curve. And the bottom radius, bottom length, half cone angle and other information of the cone target are all included in the micro-Doppler time-frequency curve of the cone bottom scattering point. The phase can be expressed as

[0099]

[0100] In the formula, is the wavelength of the narrowband radar transmission signal, is the slow time, representing the radar pulse transmission time, is the initial phase.

[0101] Step 2: Estimate the number of echo signals, determine the order of the sliding window Pole Root-MUSIC algorithm, set the window length of the sliding window PoleRoot-MUSIC algorithm, and use the sliding window to estimate the micro-Doppler time-frequency curve of the target scattering point. The target scattering point is two scattering points of the smooth precessing cone target under radar observation, namely, a sliding scattering point and a fixed scattering point.

[0102] When using the sliding window Pole Root-MUSIC algorithm to extract frequency, the number of echo signals is the same as the order that the algorithm needs to determine. Therefore, the correct estimation of the number of signals is crucial. Figure 2The number of signals in the narrowband radar echo shown in the figure is estimated to realize the order determination of the sliding window Pole Root-MUSIC algorithm. Common methods include information criterion method, eigenvalue method, matrix rank method, etc. The information criterion method estimates the number of signals by minimizing the criteria, such as Akaike Information Criterion (AIC) and Minimum Description Length (MDL), where AIC tends to select more signals, while MDL is more conservative; the eigenvalue method estimates the number of signals by analyzing the eigenvalues ​​of the covariance matrix and judging the dividing point between the signal and the noise, which is suitable for the case of high signal-to-noise ratio; the matrix rank method estimates the number of signals by singular value decomposition based on the number of singular values ​​greater than a certain threshold, which is suitable for the case of low signal-to-noise ratio; in addition, the maximum likelihood method and sparse representation method can also be used to estimate the number of signals in complex environments, but the computational complexity is high. Choosing an appropriate method will help improve the accuracy of frequency estimation. The impact of signal number determination on micro-Doppler information extraction is shown in the figure. Figure 3 and Figure 4 shown.

[0103] The window length of the sliding window Pole Root-MUSIC algorithm is generally 1 / 10 to 1 / 4 of the signal length. This step uses the sliding window Pole Root-MUSIC algorithm. According to the number of echo signals, the input echo signal estimates the instantaneous frequency of the target scattering point, that is, the sliding window estimates the micro-Doppler time-frequency curve of the target scattering point. Specifically, by eigendecomposing the signal covariance matrix and constructing an equation based on the noise subspace, the frequency information of the signal is directly obtained using its root.

[0104] Since the micro-Doppler frequency of the smoothly precessing cone target obtained by long-term observation changes sinusoidally or quasi-sinusoidally with time, the signal frequency changes rapidly. Therefore, in order to extract the micro-Doppler characteristics of the target, the echo signal needs to be divided into overlapping signal sub-segments to estimate the instantaneous micro-Doppler frequency, where the first Signal subfragment It is expressed as:

[0105]

[0106] in, It is The amplitude of the signal sub-segment, K is the number of signal sub-segments, which is related to the window length and the number of pulses. It is The instantaneous micro-Doppler frequency of the signal sub-segment, It is The instantaneous phase of a signal sub-segment, It is the additive white noise of the smooth precessing cone target echo signal.

[0107] After discretization, the signal sub-segments are discretized into points, and obtain a discrete signal ,as follows:

[0108]

[0109] In the formula, The discretization of the signal sub-segment points.

[0110] Solve for the discrete signal The covariance matrix of and the estimation matrix ,as follows:

[0111]

[0112]

[0113] Perform eigendecomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors. The eigenvectors will be used to construct the signal subspace and the noise subspace Since the signal subspace and the noise subspace are orthogonal to each other, we have:

[0114]

[0115] therefore:

[0116]

[0117] set up:

[0118]

[0119] In the formula, represents the noise power, express The identity matrix, is a Matrix, represented by The diagonal matrix composed of the eigenvalues ​​of is a The matrix consists of the eigenvectors corresponding to the signal subspace, express The conjugate transposed matrix of represents the eigenvalue belonging to the signal subspace, represents the eigenvalue belonging to the noise subspace, express The conjugate transposed matrix of express The conjugate transposed matrix of express The conjugate transposed matrix of . represents the steering vector, , yes of Power, For polynomial The roots on the unit circle are Expand and simplify to get:

[0120]

[0121] By solving the above equation, we can determine the root on the unit circle. Usually, we only select the root closest to the unit circle or the root on the circle. roots, which provide discrete signals The estimated instantaneous micro-Doppler frequency , which can be expressed as:

[0122]

[0123] Represents the solution of the equation The roots obtained, express The micro-Doppler time-frequency curve of the target scattering point is Representation.

[0124] The instantaneous frequency estimation result is as follows: Figure 5 and Figure 6 When the signal-to-noise ratio is low, a large number of outliers will appear at the intersection of the signals, such as Figure 5 As shown; when the signal-to-noise ratio is high, there will be a problem of non-crossing at the signal crossing position, such as Figure 6 These problems will make it difficult to correlate the instantaneous time-frequency analysis of the sliding window Pole Root-MUSIC algorithm, and it is necessary to remove outliers and correctly correlate the micro-Doppler time-frequency curves.

[0125] Step 3: De-noise the micro-Doppler time-frequency curve of the target scattering point.

[0126] Figure 1 The time-frequency analysis results of the echo signal with a signal-to-noise ratio of 0dB are as follows: Figure 7 and Figure 8As shown in Figure 8. The time-frequency curves extracted based on the sliding window PoleRoot-MUSIC algorithm are mostly concentrated in the middle of the STFT time-frequency curve, but there are many outliers and stray points at the intersection of the signals, as shown in Figure 8. The extracted instantaneous time-frequency curve needs to be denoised.

[0127] The denoising process of this step first needs to locate the noise position. Since the position of noise is random, using the data histogram to locate the noise may produce false noise due to improper threshold value. If only the differential value is used to locate the noise, it will not be possible to locate the situation where there is noise at the beginning and end of the sequence. Therefore, the present invention uses both the histogram method and the differential method. If the noise located by the histogram method is not located by the differential method, then the noise is false noise. If the noise located by the histogram method is also located by the differential method, then the noise is real noise. Fig. 9 The result of denoising the real noise is shown in Fig.10 , Fig.11 , Fig.12 and Fig.13 , Fig.14 , Fig.15 As shown, the two methods adopted by the present invention can accurately locate the noise position, and the time-frequency curve denoising effect is better.

[0128] For echo multi-signal processing, when the sliding window Pole Root-MUSIC algorithm is used to obtain the instantaneous time-frequency information of the signal at each moment, there is a problem of correlation between instantaneous frequencies at different moments. If the instantaneous frequencies are not effectively correlated, the frequencies of multiple signals may overlap, resulting in time-frequency aliasing. Therefore, the subsequent steps of the present invention separate the frequencies of each signal to ensure the accuracy of the time-frequency representation and the independent analysis of the signal.

[0129] Step 4: Map the denoised curve to a two-dimensional image, perform IRadon transformation on the two-dimensional image, and then locate and identify the energy concentration of the IRadon transformation, distinguish the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point, and realize scattering point identification. Then perform Radon transformation on the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point, respectively, to obtain the corresponding two trend lines.

[0130] According to the characteristic that the instantaneous micro-Doppler information of the cone model is distributed in a quasi-sine curve, the present invention uses the principle of Radon transform curve detection to associate the instantaneous micro-Doppler. The time-frequency curve data is multiple vectors, and the mapping of the time-frequency curve to the two-dimensional image can be realized by inserting multiple vectors into the same matrix.

[0131] Assume that the two-dimensional image obtained by mapping the instantaneous micro-Doppler information extracted by the sliding window Pole Root-MUSIC algorithm is represented as , then at the angle The IRadon domain is as follows

[0132]

[0133] in, represents an image in the IRadon transform domain, represents a sine curve, , is the initial coordinate determined by the initial stage of the micro-Doppler time-frequency curve, , represents the total observation time, represents the filter, the Ram-Lak filter is commonly used, and other filters such as the Shepp-Logan filter can also be selected to suppress noise. In general, The autocorrelation method can be used to estimate. , the resolution of IRadon is different.

[0134] The above formula shows that the micro-Doppler of the cone top scattering point with sinusoidal changes will be focused into a peak value, and the micro-Doppler of the cone bottom sliding scattering point with quasi-sinusoidal changes will be defocused. By training the recognition model, the cone top and cone bottom signals can be located and recognized, wherein the cone top scattering points correspond to the energy concentrated area; the cone bottom scattering points correspond to the energy dispersed area, and are symmetrically distributed about the line connecting the cone top and the center of the two-dimensional image. Finally, the recognition model obtained by training can be used to locate and recognize the cone top and cone bottom image domain signals after IRadon transformation. By way of example, the present invention uses the YOLO recognition model.

[0135] In a further embodiment of the present invention, a trend line in the form of a sine curve is obtained by performing Radon transformation on the micro-Doppler time-frequency curve of the cone top scattering point. .

[0136] For the micro-Doppler time-frequency curve of the cone bottom scattering point, Radon transform is performed to obtain two trend lines with the same amplitude and frequency in the form of sine curves. and ,Will and The weighted summation obtains a trend line that is closest to the amplitude and frequency of the micro-Doppler time-frequency curve of the cone bottom scattering point. , for example, the weight can be 0.5. Trend line and trend line These are the two corresponding trend lines obtained by the present invention.

[0137] Specifically, assuming the positioning position is , represents the location coordinates of the energy focus in the image domain after IRadon transformation, which is set center.

[0138] According to Radon transform, we can get The trend line represented is

[0139]

[0140] in is the initial phase of the trend line, which is expressed as follows:

[0141]

[0142] is the trend line amplitude, which is expressed as follows:

[0143]

[0144] when When the value detection is located in the cone top image domain, ,when When the value detection is located in the cone bottom image domain, .

[0145] By performing IRadon transformation on the instantaneous time-frequency curve extracted by the sliding window Pole Root-MUSIC algorithm, a high-precision energy distribution diagram can be obtained. Fig.16 and Fig.17 As shown. Fig.17 The energy concentration area is divided into area 1 and area 2, and area 1 and area 2 are locally enlarged, respectively. Fig.18 and 19 . Since the micro-Doppler time-frequency curve of the cone top of the smoothly precessing cone target is a sine curve, it can be concentrated into a point after IRadon transformation. Therefore, the area 1 where the energy is more concentrated is the cone top scattering point. The micro-Doppler time-frequency curve of the sliding scattering point at the bottom of the cone is a quasi-sine curve, which is mainly composed of two sine curves with the same phase, frequency and amplitude. Therefore, its IRadon transformation is manifested as energy dispersion and symmetrical distribution about the line connecting the cone top and the image center, as shown in Fig.19 As shown in area 2. Therefore, by locating and identifying the energy concentration of IRadon transform through YOLO, the micro-Doppler time-frequency curves of the cone top and cone bottom can be distinguished.

[0146] In the embodiment of the present invention, the positions of the cone top and cone bottom in the IRadon transform domain image are located by YOLO, such as Fig. 20, after Radon transformation, we get the corresponding sine curve, marked in Fig.21 In the example, since the line is consistent with the target time-frequency curve in trend, but not completely overlapped, it is called a trend line. The weighted sum of the two trend lines is performed to obtain a trend line that is closest to the cone bottom time-frequency curve.

[0147] Step 5: By using the nearest neighbor algorithm, according to the two trend lines, the micro-Doppler time-frequency curves of the target scattering points extracted by the sliding window Pole Root-MUSIC are correlated and smoothed to achieve the micro-motion feature extraction of the target.

[0148] The present invention utilizes the basic properties of Radon transform to separate the estimated but incorrectly associated time-frequency curves to ensure the accuracy of the time-frequency representation and the independent analysis of the signal. Fig. 22 and Fig.23 It can be seen that the target time-frequency curve extracted by the algorithm proposed in the present invention has a small error, can solve the problem of cross-correlation of the micro-Doppler time-frequency curves of the smooth cone precession target scattering point, and has a high extraction accuracy.

[0149] In summary, under the condition of narrow-band radar observation, the present invention extracts the micro-Doppler time-frequency curve based on the Pole Root-MUSIC algorithm, and uses Radon transform to transform the non-correlated curve into the parameter domain for data reconstruction and association, thereby completing the reconstruction of the micro-Doppler frequency of each scattering point of the target and realizing the distinction and identification of the cone top and cone bottom signals.

[0150] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for extracting micromotion and identifying scattering points of a smooth precession cone based on a narrow band, characterized in that: The steps include: Step 1: Based on narrow-band radar, an equivalent scattering point model of a smooth precessing cone target is established; Step 2, based on the equivalent scattering point model, the number of echo signals is estimated, the sliding window Pole Root-MUSIC algorithm is ordered, the window length of the sliding window Pole Root-MUSIC algorithm is set, and the micro-Doppler time-frequency curve of the target scattering point is estimated by sliding the window; the target scattering point is two scattering points of the smooth precession cone target under radar observation; Step 3, perform IRadon transformation on the micro-Doppler time-frequency curve of the target scattering point obtained by sliding window estimation, locate and identify the energy concentration of the IRadon transformation, distinguish the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point, and realize scattering point identification; Step 4, performing Radon transformation on the micro-Doppler time-frequency curves of the distinguished cone top scattering point and the cone bottom scattering point to obtain two corresponding trend lines; Step 5: Correlate and smooth the micro-Doppler time-frequency curves of the target scattering points according to the two trend lines to extract the micro-motion features of the target.

2. The method for extracting micro-motion and identifying scattering points of a narrow-band smooth precession cone according to claim 1 is characterized in that: The step 2 sets the window length of the sliding window Pole Root-MUSIC algorithm and estimates the micro-Doppler time-frequency curve of the target scattering point by sliding the window. The implementation method is as follows: dividing the echo signal into overlapping signal sub-segments; The signal sub-fragments are discretized into points, and obtain a discrete signal and its covariance matrix, which includes the true matrix and the estimation matrix ; According to the real matrix and the estimation matrix , calculate the discrete signal The estimated instantaneous micro-Doppler frequency , and according to A micro-Doppler time-frequency curve characterizing the target scattering point.

3. The method for extracting micro-motion and identifying scattered points of a narrow-band smooth precession cone according to claim 2 is characterized in that: The echo signal is divided into overlapping signal sub-segments, wherein the first Signal subfragment It is expressed as: In the formula, It is The amplitude of the signal sub-segment, K is the number of signal sub-segments, It is The instantaneous micro-Doppler frequency of the signal sub-segment, For slow time, It is The instantaneous phase of a signal sub-segment, is the additive white noise of the smooth precession cone target echo signal, j is the imaginary number sign; The discrete signal , expressed as: In the formula, The discretization of the signal sub-segment points; The real matrix , expressed as: The estimation matrix , expressed as: In the formula, is a The matrix consists of the eigenvectors corresponding to the signal subspace, express The conjugate transposed matrix of is a Matrix, represented by The diagonal matrix composed of the eigenvalues ​​of represents the noise power, express Identity matrix; represents the signal subspace, represents the noise subspace, represents the eigenvalue belonging to the signal subspace, represents the eigenvalue belonging to the noise subspace, express The conjugate transposed matrix of express The conjugate transposed matrix of ; , , represents the steering vector, , yes of Power, For polynomial roots on the unit circle; The discrete signal The estimated instantaneous micro-Doppler frequency , expressed as: In the formula, Represents the solution of the equation The roots obtained, express phase.

4. The method for extracting micro-motion and identifying scattered points of a narrow-band smooth precession cone according to claim 1 is characterized in that: In step 3, the micro-Doppler time-frequency curve of the target scattering point estimated by the sliding window is denoised and mapped into a two-dimensional image, and the two-dimensional image is subjected to IRadon transformation; The denoising is implemented as follows: The noise is located by using the histogram method and the difference method respectively. If the noise located by the histogram method is not located by the difference method, the noise is false noise. If the noise located by the histogram method is also located by the differential method, the noise is real noise, and the real noise is subjected to denoising.

5. The method for extracting micro-motion and identifying scattering points of a narrow-band smooth precession cone according to claim 4 is characterized in that: The two-dimensional image is subjected to IRadon transformation, and the angle The IRadon domain is represented as: in, represents a two-dimensional image, represents a sine curve, , is the initial coordinate determined by the initial stage of the micro-Doppler time-frequency curve, is the target precession angular frequency of the smooth precession cone, For slow time, , represents the total observation time, represents the filter, is the integration variable.

6. The method for extracting micro-motion and identifying scattering points of a narrow-band smooth precession cone according to claim 5, characterized in that: The step 3 locates and identifies the energy concentration of the IRadon transformation and distinguishes the micro-Doppler time-frequency curves of the cone top and the cone bottom. The implementation method is as follows: For different initial phases A recognition model is trained, wherein the scattering points at the top of the cone correspond to the area where energy is concentrated; the scattering points at the bottom of the cone correspond to the area where energy is dispersed, and are symmetrically distributed about the line connecting the top of the cone and the center of the two-dimensional image; the recognition model obtained by training is used to realize the positioning and recognition of the image domain signals of the top and bottom of the cone after IRadon transformation.

7. The method for extracting micro-motion and identifying scattered points of a narrow-band smooth precession cone according to claim 5, characterized in that: In step 4, Radon transform is performed on the micro-Doppler time-frequency curves of the cone top scattering point and the cone bottom scattering point respectively to obtain two corresponding trend lines, and the method is as follows: For the micro-Doppler time-frequency curve of the cone top scattering point, a trend line in the form of a sine curve is obtained by Radon transformation. ; For the micro-Doppler time-frequency curve of the cone bottom scattering point, Radon transform is performed to obtain two trend lines in the form of sine curves. and ,Will and The weighted summation obtains a trend line that is closest to the micro-Doppler time-frequency curve of the cone bottom scattering point. .

8. The method for extracting micro-motion and identifying scattering points of a narrow-band smooth precession cone according to claim 7, characterized in that: For positioning , the trend line it represents is obtained through Radon transformation: in is the initial phase of the trend line, which is expressed as follows: is the trend line amplitude, which is expressed as follows: represents the location coordinates of the energy focus in the image domain after IRadon transformation, which is set the center of; when When the value detection is located in the cone top image domain, ,when When the value detection is located in the cone bottom image domain, .

9. The method for extracting micro-motion and identifying scattered points of a narrow-band smooth precession cone according to claim 7, characterized in that: The two trend lines and have the same amplitude and frequency, and the weight is 0.

5. and Weighted summation is performed to obtain the trend line that is closest to the amplitude and frequency of the micro-Doppler time-frequency curve of the cone bottom scattering point. .

10. The method for extracting micro-motion and identifying scattering points of a narrow-band smooth precession cone according to claim 1, characterized in that: The micro-Doppler time-frequency curves of the target scattering points are associated and smoothed by a nearest neighbor algorithm.

Images