A small UAV identification method and device based on micro-Doppler feature enhancement

Abstract

The embodiment of the present application discloses a small UAV identification method and device based on micro-Doppler feature enhancement, which belongs to the field of UAV identification technology and solves the problem of low accuracy in small UAV identification. It includes: constructing a dynamic echo signal model of the rotating blades of the UAV based on the blade length and rotation frequency parameters corresponding to the UAV; extracting features of the dynamic echo signal model of the rotating blades of the UAV, and performing differential transformation based on the extracted UAV fuselage echo signal and the rotor blade echo signal; separating the micro-Doppler component based on the differential transformation, and constructing a micro-Doppler feature map based on the micro-Doppler component; matching the blade length and rotor frequency in the micro-Doppler feature map with the preset target value corresponding to the UAV, so as to enhance the micro-Doppler energy based on the matching result and obtain a feature enhancement map; constructing a convolutional neural network based on the feature enhancement map, so as to identify the UAV to be tested through the convolutional neural network.

Description

Technical Field

[0001] The present application relates to the technical field of drone identification, and in particular to a method and device for identifying small drones based on micro-Doppler feature enhancement. Background Art

[0002] Small drones usually refer to drones with a length, width and height of less than 15 cm, a mass of 10 g-500 g, an optimal operating altitude of 45-60 m, a maximum altitude of 150 m, a flight speed of 10-20 m / s, a maximum range of about 10 km and a maximum flight time of more than 2 hours.

[0003] In the prior art, small drones are identified by using Doppler spectrum with less computational effort, and its spectrum distribution can better describe the micro-Doppler characteristics caused by the rotation of drone blades. The prior art found that the spectrum distribution of small drones is related to their physical size and structure, and used the Doppler spectrum to perform identification tasks, but did not give the specific relationship between the spectrum characteristics and the dynamic structure of the drone. In addition, the prior art also analyzed the relationship between the micro-motion characteristics of small drones and the physical model of drones, and on this basis established a drone radar time domain signal model, and then analyzed the method of obtaining drone-related motion characteristics from the Doppler spectrum.

[0004] However, due to the high similarity of traditional features of low-altitude weak targets, it is difficult to effectively distinguish different types of targets, resulting in low accuracy in small UAV recognition. In addition, the rotor size of small UAVs is small, the micro-Doppler features generated are weak, and the signal-to-noise ratio is low, making it difficult for traditional methods to extract effective and identifiable micro-Doppler features, which further affects the accuracy of small UAV recognition. Summary of the invention

[0005] The embodiments of the present application provide a small UAV identification method and device based on micro-Doppler feature enhancement, which is used to solve the following technical problems: due to the high similarity of traditional features of low-altitude weak targets, it is difficult to effectively distinguish different types of targets, and the rotor size of small UAVs is small, the micro-Doppler features generated are weak, and the signal-to-noise ratio is low, so that traditional methods are difficult to extract effective and identifiable micro-Doppler features, further affecting the accuracy of small UAV identification.

[0006] The present application embodiment adopts the following technical solutions:

[0007] The embodiment of the present application provides a method for identifying a small UAV based on micro-Doppler feature enhancement. The method includes: constructing a dynamic echo signal model of the rotating blades of the UAV based on the blade length and rotation frequency parameters corresponding to the UAV; extracting features of the dynamic echo signal model of the rotating blades of the UAV, and performing differential transformation based on the extracted UAV fuselage echo signal and the rotor blade echo signal; separating the micro-Doppler component based on the differential transformation, and constructing a micro-Doppler feature map based on the micro-Doppler component; wherein the micro-Doppler feature map includes two dimensions, namely, blade length and rotor frequency; matching the blade length and rotor frequency in the micro-Doppler feature map with the preset target values ​​corresponding to the UAV, so as to enhance the micro-Doppler energy based on the matching results and obtain a feature enhancement map; constructing a convolutional neural network based on the feature enhancement map, so as to identify the UAV to be tested through the convolutional neural network.

[0008] The embodiment of the present application applies the phase difference to the drone signal, and improves the signal-to-noise ratio of the drone micro-Doppler component by coupling the energy of the drone body to the micro-Doppler component. By separating the micro-Doppler component and constructing a feature map, the influence of the blade length and rotor frequency on the echo signal is intuitively displayed, providing an important basis for the identification of the drone. By enhancing the micro-Doppler energy that matches the preset target value, the characteristics of the rotating blades of the drone are highlighted, and the recognition of the characteristics in the subsequent identification process is improved, thereby improving the accuracy of drone identification. By constructing a neural network model and training with a feature enhancement map, automatic identification of drones is achieved, and by combining the efficiency of deep learning with the accuracy of the feature enhancement map, the accuracy and efficiency of drone identification are improved.

[0009] In one implementation of the present application, a differential transformation is performed based on the extracted UAV fuselage echo signal and the rotor blade echo signal, specifically including: converting the UAV fuselage echo signal and the rotor blade echo signal into frequency domain signals based on fast Fourier transform; eliminating clutter components of the frequency domain signal through a first rectangular high-pass filter; converting the frequency domain signal from which the clutter components are eliminated into a time domain echo signal based on an inverse fast Fourier transform; and simplifying the time domain echo signal after the clutter components are eliminated based on a time delay variable, so as to perform a differential transformation based on the simplified echo signal.

[0010] In one implementation of the present application, a differential transformation is performed based on the simplified echo signal, specifically including: determining the conjugate echo and the time-delay echo corresponding to the simplified echo signal; performing a point multiplication process based on the conjugate echo and the time-delay echo to perform a second-order differential transformation on the echo signal.

[0011] In one implementation of the present application, the micro-Doppler component is separated based on differential transformation, specifically including: converting the main Doppler component to zero frequency; converting the converted echo signal into a frequency domain signal based on fast Fourier transform; performing main Doppler component elimination processing at zero frequency on the frequency domain signal through a second rectangular high-pass filter; and converting the frequency domain signal after the main Doppler component elimination processing at zero frequency into a time domain echo signal based on inverse fast Fourier transform to separate the micro-Doppler component.

[0012] In one implementation of the present application, the blade length and rotor frequency in the micro-Doppler characteristic map are matched with the preset true values ​​corresponding to the UAV, so as to enhance the micro-Doppler energy based on the matching results to obtain a characteristic enhancement map, which specifically includes: determining a preset target rotor size range and a preset target rotor frequency range; constructing a two-dimensional search compensation factor based on the preset target rotor size range and the preset target rotor frequency range; based on the two-dimensional search compensation factor, traversing the rotor size and rotor frequency, transforming the differential transformation result to the rotor parameter plane, and obtaining the characteristic enhancement map by determining the best matching position of the signal.

[0013] In one implementation of the present application, obtaining a feature enhancement map specifically includes:

[0014] Function-based:

[0015] ;

[0016] Get the feature enhancement map; where, To preset the target rotor size range; To preset the target rotor frequency range; is the two-dimensional search compensation factor; To transform the difference transformation result to the rotor parameter plane, s is the echo signal after differential transformation.

[0017] In one implementation of the present application, the two-dimensional search compensation factor is:

[0018] ;

[0019] in, To preset the target rotor size range; To preset the target rotor frequency range; is the two-dimensional search compensation factor; is the signal delay parameter; is the slow time dimension time variable; is the radar signal wavelength.

[0020] In one implementation of the present application, a convolutional neural network is constructed based on a feature enhancement map, specifically including: extracting features from the feature enhancement map through a convolutional neural network; determining the feature extraction results corresponding to the four residual modules in the convolutional neural network; linearly mapping the feature extraction results through a fully connected layer to obtain a vector of the total number of categories, and obtaining a recognition result through a Softmax classification layer; and determining that the convolutional neural network training is completed when the recognition result meets the preset conditions.

[0021] In one implementation of the present application, the feature extraction results corresponding to the four residual modules in the convolutional neural network are determined, specifically including: convolving the input x1 of the first convolutional block based on the first convolutional layer of the first convolutional block to obtain the feature F 11 ; The features obtained from the first convolutional layer based on the nonlinear activation function F 11 Perform nonlinear mapping to obtain features F 12 , the feature F 12 Input to the second convolutional layer, through the second convolutional layer to feature F 12 Perform convolution to obtain features F 13 ; The feature F 13 Add it to the input x1 of the first convolution block to get the output of the first convolution block; convolve the output of the first convolution block based on the first convolution layer of the second convolution block to get the feature F 14 ; The features obtained by the first convolution layer of the second convolution block based on the nonlinear activation function F 14 Perform nonlinear mapping to obtain features F 15 , and then the obtained features F 15 Input to the second convolutional layer of the second convolutional block, and use the second convolutional layer of the second convolutional block to perform feature F 15 Perform convolution to obtain features F 16 The features obtained F 16 Add it to the output of the first convolutional block to get the feature extraction result of the first residual block F 1; Continue to extract features from other residual blocks until feature extraction results corresponding to the four residual modules are obtained.

[0022] The embodiment of the present application provides a small drone identification device based on micro-Doppler feature enhancement, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: construct a dynamic echo signal model of the drone rotating blades based on the blade length and rotation frequency parameters corresponding to the drone; perform feature extraction on the dynamic echo signal model of the drone rotating blades, and perform differential transformation on the extracted drone fuselage echo signal and the rotor blade echo signal; separate the micro-Doppler component based on the differential transformation, and construct a micro-Doppler feature map based on the micro-Doppler component; wherein the micro-Doppler feature map includes two dimensions of blade length and rotor frequency; match the blade length and rotor frequency in the micro-Doppler feature map with the preset target value corresponding to the drone, so as to enhance the micro-Doppler energy based on the matching result and obtain a feature enhancement map; and construct a convolutional neural network based on the feature enhancement map to identify the drone to be tested through the convolutional neural network.

[0023] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: The embodiments of the present application apply the phase difference to the drone signal, and improve the signal-to-noise ratio of the drone micro-Doppler component by coupling the energy of the drone body to the micro-Doppler component. By separating the micro-Doppler component and constructing a feature map, the influence of the blade length and rotor frequency on the echo signal is intuitively displayed, providing an important basis for the identification of the drone. By enhancing the micro-Doppler energy that matches the preset target value, the characteristics of the rotating blades of the drone are highlighted, and the recognition of the characteristics in the subsequent identification process is improved, thereby improving the accuracy of drone identification. By constructing a neural network model and training with a feature enhancement map, automatic identification of drones is achieved, and by combining the high efficiency of deep learning and the accuracy of the feature enhancement map, the accuracy and efficiency of drone identification are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0025] Figure 1 A flow chart of a small UAV identification method based on micro-Doppler feature enhancement provided in an embodiment of the present application;

[0026] Figure 2 A dynamic geometric model of a small quad-rotor drone provided in an embodiment of the present application;

[0027] Figure 3 A flow chart of a micro-Doppler feature enhancement algorithm provided in an embodiment of the present application;

[0028] Figure 4 A spectrum diagram of a drone signal before zero-frequency clutter is removed provided in an embodiment of the present application;

[0029] Figure 5 A spectrum diagram of a drone signal after removing zero-frequency clutter provided in an embodiment of the present application;

[0030] Figure 6 A spectrum diagram of a drone signal after phase difference before removing the main Doppler component provided in an embodiment of the present application;

[0031] Figure 7 A spectrum diagram of a drone signal after phase difference after removing the main Doppler component provided in an embodiment of the present application;

[0032] Figure 8 A UAV micro-Doppler enhancement characteristic map provided in an embodiment of the present application;

[0033] Fig. 9 A CNN framework diagram constructed in accordance with an embodiment of the present application;

[0034] Fig.10 A first schematic diagram of an algorithm recognition result provided in an embodiment of the present application;

[0035] Fig.11 A second schematic diagram of an algorithm recognition result provided in an embodiment of the present application;

[0036] Fig.12 A third schematic diagram of an algorithm recognition result provided in an embodiment of the present application;

[0037] Fig.13 A schematic diagram of the structure of a small UAV identification device based on micro-Doppler feature enhancement provided in an embodiment of the present application.

[0038] Reference numerals:

[0039] 200 Small UAV identification device based on micro-Doppler feature enhancement, 201 processor, 202 memory. DETAILED DESCRIPTION

[0040] The embodiments of the present application provide a method and device for identifying a small UAV based on micro-Doppler feature enhancement.

[0041] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0042] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0043] Figure 1 A flowchart of a small drone identification method based on micro-Doppler feature enhancement provided in an embodiment of the present application is shown in FIG. Figure 1 As shown in FIG. 1 , the small UAV identification method based on micro-Doppler feature enhancement includes the following steps:

[0044] Step 101: construct a dynamic echo signal model of the rotating blades of the UAV based on the blade length and rotation frequency parameters corresponding to the UAV.

[0045] In one embodiment of the present application, a far-field motion model of the UAV is first constructed based on the dynamics of the rotation of the UAV blades.

[0046] Specifically, the embodiment of the present application performs modeling and analysis on a small UAV with a typical quad-rotor and two blades on each rotor shaft. Figure 2 A dynamic geometric model of a small quad-rotor drone provided in an embodiment of the present application, such as Figure 2 As shown in the figure, a long-range signal model of a long-range UAV is established. The radar is located at the origin O of the Cartesian coordinate system, and the distance between the UAV geometric center and the radar is defined as , the angle between the connecting line and the XY plane is Assuming that the radar coherent processing interval is short enough, the translation and rotor micromotion of the UAV can be expressed as constant radial velocity v and constant angular velocity To model, define the translation speed of the drone as v , No. p The angular velocity of the rotor is If the UAV geometric center and the p The distance between the centers of the rotors is , the angle is , then the radar and the p The distance between the rotor centers It can be expressed as follows:

[0047] ;

[0048] Among them, since the UAV target is in the far field of the radar, there are ,therefore It can be approximately expressed as:

[0049] ;

[0050] It is only related to the translation of the UAV. Assume that the first q Scatterers in p The distance between the radar and the center of the drone rotor blade is defined on the rotor blade. for:

[0051] ;

[0052] in, yes The angle between the XY plane and The first q The scatterer and p The distance between the rotor centers, It is q The initial phase of a scatterer is , so there is , then the distance between the radar and the center of the UAV rotor blade is It can be approximately expressed as:

[0053] ;

[0054] in, R Mpq ( t ) Micro-Doppler component in the signal model.

[0055] In one embodiment of the present application, a radar signal model of a drone is constructed.

[0056] Specifically, the first q The distance change of a scatterer can be expressed as the linear superposition of translation and micromotion. N The rotor blades N The leaves have N Different initial phase angles ,in The time domain signal model of the UAV target is expressed as:

[0057] ;

[0058] in, K is the number of scatterers on the main part of the small drone, It is k The complex amplitude of the scatterer, Pis the number of rotors, Q is the number of scatterers per leaf, It is q The first rotor q Substituting the variables mentioned above into the above formula and merging the constant terms, the UAV signal model can be simplified to the linear superposition of the UAV body translation and the rotor translation + micro-motion:

[0059] ;

[0060] in:

[0061] ;

[0062] ;

[0063] ;

[0064] in, , , All are intermediate variables; is the distance between the UAV geometric center and the radar; is the geometric center of the UAV and k The distance between the centers of the rotors; the spectrum of the small UAV can be obtained by Fourier transform, expressed as:

[0065] ;

[0066] in Yes The Fourier transform result of is expressed as:

[0067] ;

[0068] We can further use the first category k Bessel function To indicate that there are:

[0069] ;

[0070] in:

[0071] ;

[0072] ;

[0073] ;

[0074] Therefore, Using Fourier series decomposition, it can be expressed as follows:

[0075] ;

[0076] Furthermore, Perform Fourier transform to get its spectrum The expression is:

[0077] ;

[0078] in, It's Dirac δ Function. Substituting in, we can get The expression is:

[0079] ;

[0080] in, u is an intermediate variable; Different parameters caused by different scattering centers B The mean value of k no N When the integer multiple of , the above formula is approximately 0, so it can be further simplified to:

[0081] ;

[0082] By integrating the above formula, we can finally get the spectrum expression of the small UAV:

[0083] ;

[0084] From the above formula, we can see that the spectrum of the small drone consists of two parts: the center frequency spectrum caused by the main body movement, ; and around the main Doppler, pairs of harmonic lines caused by rotor rotation, In the field of UAV rotor frequency research, the interval frequency of these harmonic spectral lines is usually called the chopping frequency, which is expressed as follows:

[0085] ;

[0086] in, N is the number of rotor blades, For the p The angular velocity of the rotor.

[0087] Step 102: extract features from the dynamic echo signal model of the UAV rotating blades, and perform differential transformation on the extracted UAV fuselage echo signal and the rotor blade echo signal.

[0088] Figure 3 A flow chart of a micro-Doppler feature enhancement algorithm provided in an embodiment of the present application is as follows: Figure 3 As shown, the micro-Doppler feature enhancement algorithm in the embodiment of the present application includes obtaining drone echo, clutter component elimination, phase difference, main Doppler component elimination and two-dimensional search compensation, so as to obtain a micro-Doppler feature enhancement map.

[0089] In one embodiment of the present application, the drone fuselage echo signal and the rotor blade echo signal are converted into frequency domain signals based on fast Fourier transform. The frequency domain signal is subjected to clutter component elimination through a first rectangular high-pass filter. Based on inverse fast Fourier transform, the frequency domain signal from which the clutter component is eliminated is converted into a time domain echo signal. Based on a time delay variable, the time domain echo signal from which the clutter component is eliminated is simplified, so as to perform a differential transformation based on the simplified echo signal.

[0090] Specifically, due to the large beam width of the radar used, the collected low-elevation airborne target data contains strong ground clutter. Therefore, after obtaining the radar echo signal of the drone, zero-frequency clutter removal must be performed to prevent the clutter component from affecting the performance of subsequent algorithms. First, the time domain echo of the drone is processed using the fast Fourier transform. Transform to obtain its frequency domain signal Next, the first rectangular high-pass filter is designed H ( f ) to eliminate the zero-frequency clutter component, which is expressed as follows:

[0091] ;

[0092] ;

[0093] in, It is a frequency threshold parameter determined based on the statistical analysis of the Doppler spread of clutter in the measurement data. By applying the operation in the above formula, the frequency domain signal of the drone is multiplied element by element by the high-pass filter to obtain the frequency domain signal with the clutter component removed. The frequency domain results before and after the zero-frequency clutter component is removed are obtained. Finally, the zero-frequency clutter component is removed by inverse fast Fourier transform. The frequency domain signal is converted to obtain its time domain signal to be input into the next module.

[0094] Figure 4 A spectrum diagram of a drone signal before removing zero-frequency clutter is provided in an embodiment of the present application, such as Figure 4 As shown, the horizontal axis is the Doppler frequency and the vertical axis is the frequency domain echo amplitude. Figure 5 A spectrum diagram of a drone signal after removing zero-frequency clutter is provided in an embodiment of the present application, such as Figure 5As shown, the horizontal axis is the Doppler frequency, and the vertical axis is the frequency domain echo amplitude. Figure 4 and Figure 5 By comparison, it can be seen that the clutter is significantly reduced after removal.

[0095] In one embodiment of the present application, the conjugate echo and the time-delay echo corresponding to the simplified echo signal are determined, and a point multiplication process is performed based on the conjugate echo and the time-delay echo to perform a second-order difference transformation on the echo signal.

[0096] Specifically, for the drone echo signal s ( t ) is the discrete variable slow time t complex-valued function, introducing the delay variable Δ t , considering the initial phase randomness of the UAV frequency in the actual signal, simplify the modeling formula, the UAV echo Defined as:

[0097] ;

[0098] in and are the average amplitudes of the drone body and rotor, respectively. The variance is Gaussian white noise. and is defined as follows:

[0099] ;

[0100] Among them, the change of the drone's Doppler speed causes the micro-Doppler energy to diverge, resulting in low integration efficiency of the micro-Doppler component, making it difficult to achieve a sufficient signal-to-noise ratio for drone target identification. The embodiment of the present application draws on the concept of the discrete polynomial phase transformation algorithm and applies the phase difference to the drone signal. By coupling the energy of the drone body to the micro-Doppler component, the goal is to improve the signal-to-noise ratio of the drone's micro-Doppler component.

[0101] Furthermore, the drone echo is transformed by a second-order difference, and the conjugate echo of the drone echo is transformed into Delayed Echo Point multiplication, the transformation result is:

[0102] ;

[0103] in, is the noise component after phase difference. , , and It is expressed as follows:

[0104] ;

[0105] ;

[0106] in The original noise The conjugate component of for The delay component of It is an intermediate iteration variable.

[0107] It can be seen from the above formula that after the differential transformation, the transformation results have a total of three items. The embodiment of the present application moves most of the main Doppler components to zero frequency, which is the first item in the formula. The main Doppler components can be removed by setting the components at zero frequency to zero in the frequency domain. The last two items in the formula are both harmonic components and can be used as subsequent compensation factor items. It should be noted that the second item in the formula adds an amplitude modulation of the drone body to the harmonic component. A 1, further improving the signal-to-noise ratio of the harmonic components of the drone, so the subsequent compensation algorithm of the embodiment of the present application selects the second term of the differential transformation as the compensation factor.

[0108] Step 103: Separate the micro-Doppler component based on the differential transformation, and construct a micro-Doppler characteristic map according to the micro-Doppler component.

[0109] In one embodiment of the present application, the main Doppler component is converted to zero frequency. Based on fast Fourier transform, the converted echo signal is converted into a frequency domain signal. The frequency domain signal is subjected to main Doppler component elimination processing at zero frequency through a second rectangular high-pass filter. Based on inverse fast Fourier transform, the frequency domain signal after the main Doppler component elimination processing at zero frequency is converted into a time domain echo signal to separate the micro-Doppler component. The micro-Doppler characteristic diagram includes two dimensions: blade length and rotor frequency.

[0110] Figure 6 A spectrum diagram of a drone signal after phase difference before removing the main Doppler component provided in an embodiment of the present application, such as Figure 6 As shown, the horizontal axis is frequency and the vertical axis is amplitude. Figure 7 A spectrum diagram of a drone signal after phase difference after removing the main Doppler component provided in an embodiment of the present application, such as Figure 7 As shown, the horizontal axis is frequency and the vertical axis is amplitude. Figure 6 and Figure 7 By comparison, it can be seen that the main Doppler component is significantly reduced after removal.

[0111] Specifically, considering the subsequent two-dimensional exhaustive search and compensation operations, it is necessary to first determine the impact of the first and third items of the final result of the previous step on the subsequent steps. The third item can be ignored. For the first item, if it is introduced in the subsequent operation, it is a constant component with a very high amplitude. It will increase the background noise of the entire compensation map, which is not conducive to the search for micro-Doppler components. Therefore, it is necessary to remove the main Doppler component that has been offset to zero frequency.

[0112] Further, the signal is transformed into the frequency domain using a fast Fourier transform. Subsequently, the frequency domain signal is element-wise multiplied by a second rectangular high-pass filter to eliminate the main Doppler component at zero frequency. Finally, the frequency domain signal is converted back to the time domain using an inverse fast Fourier transform operation to obtain a time domain signal suitable for input into subsequent modules.

[0113] Step 104: Match the blade length and rotor frequency in the micro-Doppler characteristic map with the preset target values ​​corresponding to the UAV, so as to enhance the micro-Doppler energy based on the matching result and obtain a characteristic enhancement map.

[0114] In one embodiment of the present application, a preset target rotor size range and a preset target rotor frequency range are determined. Based on the preset target rotor size range and the preset target rotor frequency range, a two-dimensional search compensation factor is constructed. Based on the two-dimensional search compensation factor, the rotor size and rotor frequency are traversed, and the differential transformation result is transformed into a rotor parameter plane, and a feature enhancement map is obtained by determining the best matching position of the signal.

[0115] Specifically, in order to obtain a micro-Doppler signature with a higher signal-to-noise ratio, the embodiment of the present application designs a two-dimensional search compensation factor based on blade and rotor speed. An exhaustive search is performed in the two-dimensional plane to determine the position that best matches the signal, thereby improving the signal-to-noise ratio of the drone micro-Doppler signature.

[0116] Traditional search compensation algorithms are usually used for multi-dimensional joint search of target motion trajectories in fast / slow time matrix data collected by radar. The embodiment of the present application extends the application of search compensation to the drone identification task.

[0117] Specifically, based on the phase difference results, after removing the body Doppler component at zero frequency, two micro-Doppler components are retained. The first term represents the coupling between the drone body Doppler component and the rotor micro-Doppler, while the second term corresponds to the rotor micro-Doppler component. Considering that the drone body is much larger than the rotor, the energy of the rotor micro-motion component in the coupling term is improved and considered as a component for subsequent compensation. The ranges of the target rotor size and rotor frequency are set to be and , giving the compensation factor The expression is:

[0118] ;

[0119] In actual algorithm use, in order to remove the impact of compensation factor amplitude changes on compensation, the actual compensation factor is as follows:

[0120] ;

[0121] in, To preset the target rotor size range; To preset the target rotor frequency range; is the two-dimensional search compensation factor; is the signal delay parameter; is the slow time dimension time variable; is the radar signal wavelength.

[0122] Finally, the rotor size and rotor frequency are traversed, and the difference results are transformed to the rotor parameter plane. The transformation results are as follows:

[0123] ;

[0124] The feature enhancement map obtained by transformation. Among them, To preset the target rotor size range; To preset the target rotor frequency range; is the two-dimensional search compensation factor; It is the transformation result after transforming the difference transformation result to the rotor parameter plane.

[0125] The point with the highest energy corresponds to the location where the target matches the compensation parameters most closely. The compensated drone frequency shows a gradually weakening dotted line characteristic. In addition, the enhanced feature map The noise is defined as follows:

[0126] ;

[0127] in, is the noise component after phase difference; To enhance the noise of feature maps; is the compensation factor for two-dimensional search.

[0128] Figure 8 A UAV micro-Doppler enhancement feature map provided in an embodiment of the present application, such as Figure 8 As shown, the horizontal axis is the rotation frequency of the UAV rotor, and the vertical axis is the compensation of the UAV blade length. It can be seen from the figure that on the newly generated enhanced feature map, there is a strong point distribution component at the corresponding UAV rotor rotation frequency, which is the micro-Doppler characteristic component of the UAV.

[0129] Step 105: construct a convolutional neural network based on the feature enhancement graph to identify the drone to be tested through the convolutional neural network.

[0130] In one embodiment of the present application, a feature enhancement graph is extracted through a convolutional neural network, and the feature extraction results corresponding to the four residual modules in the convolutional neural network are determined. The feature extraction results are linearly mapped through a fully connected layer to obtain a vector of the total number of categories, and the recognition result is obtained through a Softmax classification layer. When the recognition result meets the preset conditions, it is determined that the convolutional neural network training is completed.

[0131] Specifically, the embodiment of the present application uses a convolutional neural network to extract deep features of the target enhanced feature map. Fig. 9 A CNN framework diagram constructed in an embodiment of the present application is shown in FIG. Fig. 9 As shown in the figure, the input target enhanced feature map is passed through the convolution layer and the pooling layer to obtain the extracted deep features.

[0132] Specifically, in the first step, the feature extraction results of four residual blocks are calculated. The four residual blocks are the first residual block, the second residual block, the third residual block, and the fourth residual block. The feature extraction result of the first residual block is F 1. The feature extraction result of the second residual block is F 2. The feature extraction result of the third residual block is F 3. The feature extraction result of the fourth residual block is F 4.

[0133] Among them, the two convolution blocks included in the first residual block are the first convolution block and the second convolution block. The feature extraction result of the first residual block is F 1 is obtained by:

[0134] (1) Use the first convolutional layer of the first convolutional block to convolve the input x1 of the first convolutional block to obtain the feature ,in, L The number of channels representing features, express F The i channels, i =1, 2, …, L , It is expressed by the following formula:

[0135] ;

[0136] in, represents the convolution operation, Representative i convolution kernels, Representativesi The offset of the convolution kernel;

[0137] (2) Use nonlinear activation function to obtain the features of the first convolutional layer F 11 Perform nonlinear mapping to obtain features F 12 , and then the obtained features F 12 Input to the second convolutional layer, use the second convolutional layer to feature F 12 Perform convolution to obtain features F 13 ;

[0138] (3) The features obtained in step (2) F 13 Add it to the input x1 of the first convolution block to get the output of the first convolution block.

[0139] Similarly, the method for obtaining the output of the second convolutional block in the first residual block is:

[0140] (4) Use the first convolutional layer of the second convolutional block to convolve the output of the first convolutional block to obtain the feature ,in, L The number of channels representing features, express F The i channels, i =1, 2, …, L , It is expressed by the following formula:

[0141] ;

[0142] in, represents the convolution operation, Representative i convolution kernels, Representatives i The offset of the convolution kernel;

[0143] (5) Use the nonlinear activation function to obtain the features of the first convolution layer of the second convolution block F 14 Perform nonlinear mapping to obtain features F 15 , and then the obtained features F 15 Input to the second convolutional layer of the second convolutional block, and use the second convolutional layer of the second convolutional block to perform feature F 15 Perform convolution to obtain features F 16 ;

[0144] (6) The features obtained in step (5) F 16 Added to the output of the first convolutional block, the feature extraction result of the first residual block is F 1;

[0145] The feature extraction result of the first residual block is F 1 is used as the input of the first convolution block in the second residual block, and the method of steps (1)-(3) is adopted to obtain the output of the first convolution block in the second residual block. Then the output of the first convolution block in the second residual block is used as the input of the second convolution block in the second residual block, and the method of steps (4)-(6) is adopted to obtain the feature extraction result of the second residual block: F 2;

[0146] The feature extraction result of the second residual block is F 2 is used as the input of the first convolution block in the third residual block, and the method of steps (1)-(3) is adopted to obtain the output of the first convolution block in the third residual block. Then the output of the first convolution block in the third residual block is used as the input of the second convolution block in the third residual block, and the method of steps (4)-(6) is adopted to obtain the feature extraction result of the third residual block: F 3;

[0147] The feature extraction result of the third residual block is F 3 is used as the input of the first convolution block in the fourth residual block, and the method of steps (1)-(3) is adopted to obtain the output of the first convolution block in the fourth residual block. Then the output of the first convolution block in the fourth residual block is used as the input of the second convolution block in the fourth residual block, and the method of steps (4)-(6) is adopted to obtain the feature extraction result of the fourth residual block: F 4.

[0148] The second step is to build a classification module:

[0149] Finally, the features are linearly mapped through the fully connected layer to obtain a vector of the total number of categories. , enter the Softmax classification layer, and get the recognition result. i The probability that a target belongs to each category is as follows:

[0150] ;

[0151] in, c Represents the category number, C Represents the total number of categories, and calculates the target in sequence i Probability of belonging to each category Then, the maximum value is the category to which the target belongs.

[0152] Fig.10 A first schematic diagram of an algorithm recognition result provided in an embodiment of the present application, Fig.11 A second schematic diagram of an algorithm recognition result provided in an embodiment of the present application, Fig.12 The third schematic diagram of an algorithm recognition result provided in an embodiment of the present application. Fig.10 , Fig.11 and Fig.12 As shown in the figure, the horizontal axis is frequency and the vertical axis is amplitude. It can be seen from the schematic diagram after recognition that the deep features of the target enhanced feature map can be accurately extracted by constructing a convolutional neural network.

[0153] In order to verify the target wing flapping pattern recognition method described above, a small UAV identification method based on micro-Doppler feature enhancement described in an embodiment of the present application is adopted to complete the pattern recognition of 8447 groups of target echo data sets based on experimental measurements (3119 groups of UAVs and 5328 groups of other targets). If the verification is passed, the method is determined to be feasible, and the convolutional neural network constructed based on the feature enhancement graph is used to identify the UAV to be tested.

[0154] Fig.13 A schematic diagram of the structure of a small drone identification device based on micro-Doppler feature enhancement provided in an embodiment of the present application. Fig.13 As shown, a small UAV identification device 200 based on micro-Doppler feature enhancement includes: at least one processor 201; and a memory 202 that is communicatively connected to the at least one processor 201; wherein the memory 202 stores instructions that can be executed by the at least one processor 201, and the instructions are executed by the at least one processor 201 so that the at least one processor 201 can: construct a dynamic echo signal model of the rotating blades of the UAV based on the blade length and rotation frequency parameters corresponding to the UAV; perform feature extraction on the dynamic echo signal model of the rotating blades of the UAV, and perform differential transformation on the extracted UAV fuselage echo signal and the rotor blade echo signal; based on the differential transformation, separate the micro-Doppler component, and construct a micro-Doppler feature map according to the micro-Doppler component; wherein the micro-Doppler feature map includes two dimensions of blade length and rotor frequency; match the blade length and rotor frequency in the micro-Doppler feature map with the preset target value corresponding to the UAV, so as to enhance the micro-Doppler energy based on the matching result to obtain a feature enhancement map; and construct a convolutional neural network based on the feature enhancement map to identify the UAV to be tested through the convolutional neural network.

[0155] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0156] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the embodiments of the present application may have various changes and variations. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A small UAV identification method based on micro-Doppler feature enhancement, characterized in that: The method comprises: Based on the blade length and rotation frequency parameters corresponding to the UAV, a dynamic echo signal model of the UAV rotating blades is constructed; Extracting features from the dynamic echo signal model of the UAV rotating blades, and performing differential transformation on the extracted UAV fuselage echo signal and the rotor blade echo signal; Based on the differential transformation, a micro-Doppler component is separated, and a micro-Doppler characteristic map is constructed according to the micro-Doppler component; wherein the micro-Doppler characteristic map includes two dimensions: blade length and rotor frequency; Matching the blade length and rotor frequency in the micro-Doppler characteristic map with preset target values ​​corresponding to the UAV, so as to enhance the micro-Doppler energy based on the matching result and obtain a characteristic enhancement map; A convolutional neural network is constructed based on the feature enhancement graph to identify the unmanned aerial vehicle to be tested through the convolutional neural network.

2. According to claim 1, a method for identifying small UAVs based on micro-Doppler feature enhancement is characterized in that: The step of performing differential transformation on the extracted UAV fuselage echo signal and the rotor blade echo signal specifically includes: The UAV fuselage echo signal and the rotor blade echo signal are converted into frequency domain signals based on fast Fourier transform; Eliminating clutter components of the frequency domain signal through a first rectangular high-pass filter; Based on inverse fast Fourier transform, the frequency domain signal with clutter components eliminated is converted into a time domain echo signal; Based on the time delay variable, the time domain echo signal after clutter component elimination is simplified, so as to perform differential transformation based on the simplified echo signal.

3. The method for identifying small UAVs based on micro-Doppler feature enhancement according to claim 2 is characterized in that: The performing differential transformation based on the simplified echo signal specifically includes: Determine the conjugate echo and the time-delay echo corresponding to the simplified echo signal; Point multiplication is performed based on the conjugate echo and the time-delay echo to perform a second-order difference transformation on the echo signal.

4. The method for identifying small UAVs based on micro-Doppler feature enhancement according to claim 1, characterized in that: The step of separating the micro-Doppler component based on the differential transformation specifically includes: Convert the main Doppler component to zero frequency; Based on fast Fourier transform, converting the converted echo signal into a frequency domain signal; Performing a process of eliminating a main Doppler component at a zero frequency on the frequency domain signal through a second rectangular high-pass filter; Based on inverse fast Fourier transform, the frequency domain signal after the main Doppler component at zero frequency is eliminated is converted into a time domain echo signal to separate the micro-Doppler component.

5. The method for identifying small UAVs based on micro-Doppler feature enhancement according to claim 1, characterized in that: The blade length and rotor frequency in the micro-Doppler characteristic map are matched with preset real values ​​corresponding to the UAV to enhance the micro-Doppler energy based on the matching result to obtain a characteristic enhancement map, specifically including: Determining a preset target rotor size range and a preset target rotor frequency range; Constructing a two-dimensional search compensation factor based on the preset target rotor size range and the preset target rotor frequency range; Based on the two-dimensional search compensation factor, the rotor size and rotor frequency are traversed, the differential transformation result is transformed to the rotor parameter plane, and the feature enhancement map is obtained by determining the best matching position of the signal.

6. The method for identifying small UAVs based on micro-Doppler feature enhancement according to claim 5, characterized in that: The obtaining of the feature enhancement map specifically includes: Function-based: ； The feature enhancement map is obtained; wherein, To preset the target rotor size range; To preset the target rotor frequency range; is the two-dimensional search compensation factor; To transform the difference transformation result to the rotor parameter plane, s is the echo signal after differential transformation.

7. The method for identifying small UAVs based on micro-Doppler feature enhancement according to claim 6, characterized in that: The two-dimensional search compensation factor is: ； in, To preset the target rotor size range; To preset the target rotor frequency range; is the two-dimensional search compensation factor; is the signal delay parameter; is the slow time dimension time variable; is the wavelength of the radar signal.

8. The method for identifying small UAVs based on micro-Doppler feature enhancement according to claim 1, characterized in that: The constructing of a convolutional neural network based on the feature enhancement graph specifically includes: Performing feature extraction on the feature enhancement map through a convolutional neural network; Determine the feature extraction results corresponding to the four residual modules in the convolutional neural network respectively; The feature extraction result is linearly mapped through a fully connected layer to obtain a vector of the total number of categories, and then the recognition result is obtained through a Softmax classification layer; When the recognition result meets the preset conditions, it is determined that the convolutional neural network training is completed.

9. The method for identifying small UAVs based on micro-Doppler feature enhancement according to claim 8, characterized in that: Determining the feature extraction results corresponding to the four residual modules in the convolutional neural network specifically includes: Based on the first convolutional layer of the first convolutional block, the input x1 of the first convolutional block is convolved to obtain the feature F 11 ; Based on the nonlinear activation function, the features obtained by the first convolutional layer F 11 Perform nonlinear mapping to obtain features F 12 , the features F 12 Input to the second convolutional layer, through which the features F 12 Perform convolution to obtain features F 13 ; The characteristics F 13 Add it to the first convolution block input x1 to obtain the output of the first convolution block; Based on the first convolution layer of the second convolution block, the output of the first convolution block is convolved to obtain the feature F 14 ; The features obtained by the first convolutional layer of the second convolutional block based on the nonlinear activation function F 14 Perform nonlinear mapping to obtain features F 15 , and then the obtained features F 15 Input to the second convolutional layer of the second convolutional block, and use the second convolutional layer of the second convolutional block to perform feature F 15 Perform convolution to obtain features F 16 ; The features obtained F 16 Add it to the output of the first convolutional block to get the feature extraction result of the first residual block F 1; Continue to perform feature extraction on other residual blocks until feature extraction results corresponding to the four residual modules are obtained.

10. A small drone identification device based on micro-Doppler feature enhancement, characterized in that: The device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute the method according to any one of claims 1 to 9.

Images