Low-altitude target detection method and device, computer device and readable storage medium

Abstract

The application relates to a low-altitude target detection method and device, computer equipment and a readable storage medium, comprising: obtaining a micro-Doppler image corresponding to an echo signal of a to-be-identified low-altitude target; identifying the micro-Doppler image by using a target image fingerprint library to obtain category information of the to-be-identified low-altitude target; the target image fingerprint library is obtained by a target residual neural network; if the category information is a drone, estimating a drone state parameter of the to-be-identified low-altitude target, wherein the drone state parameter comprises a blade number, a distance and a speed. The application uses a residual network to solve the gradient vanishing problem of a deep network, constructs a target image fingerprint library, realizes rapid identification of low-altitude target type information, synchronously completes distance, speed and other drone state parameter estimation, supports low-altitude global integrated perception, and improves the identification accuracy of a low-altitude target drone in a complex environment.

Description

Technical Field

[0001] This application relates to the field of low-altitude target detection and intelligent identification technology, and in particular to a low-altitude target detection method, apparatus, computer equipment, and readable storage medium. Background Technology

[0002] With increasingly frequent low-altitude airspace activities, rotary-wing drones are widely used in logistics inspection, aerial surveying and mapping, and emergency rescue due to their maneuverability, low cost, and ease of operation. However, at the same time, the risks of illegal intrusion, spying on sensitive areas, and interference / sabotage are also increasing. Targets in the low-altitude environment are typically characterized by low-altitude flight, slow speed, and small size, and are often accompanied by strong static clutter interference from buildings and ground facilities. This causes traditional detection methods to face problems such as limited detection range, high false alarm rate, and insufficient target differentiation capabilities.

[0003] Among related technologies, methods for identifying aerial targets based on micro-Doppler features generally suffer from problems such as the micro-Doppler features not being easily highlighted in strong clutter backgrounds, insufficient robustness in low-altitude complex environments, and insufficient integration of identification structure and parameter estimation. Summary of the Invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, and readable storage medium for detecting, identifying, and analyzing targets with micro-motion characteristics such as birds and drones in low-altitude airspace, which can reliably sense and efficiently identify low-altitude targets in cluttered environments.

[0005] In a first aspect, this application provides a method for detecting low-altitude targets, including:

[0006] Acquire the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified;

[0007] The micro-Doppler image is identified using a target image fingerprint database to obtain the category information of the low-altitude target to be identified; the target image fingerprint database is constructed through a target residual neural network.

[0008] If the category information is a drone, estimate the drone state parameters of the low-altitude target to be identified, wherein the drone state parameters include the number of blades, distance, and speed.

[0009] In one embodiment, the method further includes:

[0010] Acquire composite echo signals of low-altitude targets, wherein the composite echo signals include static clutter;

[0011] The Pratt circle fitting algorithm is used to perform static clutter suppression on the composite echo signal to obtain the target echo signal after clutter suppression.

[0012] A smooth pseudo-Wegener-Weil distribution was used to perform time-frequency analysis on the target echo signal to obtain a micro-Doppler time-frequency image;

[0013] The target residual neural network is trained using the micro-Doppler time-frequency image, and the target image fingerprint database is constructed based on the target residual neural network.

[0014] In one embodiment, the step of using the Pratt circle fitting algorithm to perform static clutter suppression on the composite echo signal to obtain the clutter-suppressed target echo signal includes:

[0015] Extract the single-carrier slow time sequence of the composite echo signal and map the single-carrier slow time sequence to the IQ complex plane to obtain the original slow time sequence;

[0016] A Pratt algebraic circle equation model is constructed, and the generalized eigenvalue problem is solved by combining the constraints to obtain the coordinates of the static clutter center.

[0017] The clutter component corresponding to the static clutter center coordinate is removed from the original slow time series, and the Doppler rotation component is translated to the origin of the complex plane to complete the clutter suppression of the composite echo signal, thereby obtaining the target echo signal.

[0018] In one embodiment, the step of performing time-frequency analysis on the target echo signal using a smooth pseudo-Wigner-Weil distribution to obtain a micro-Doppler time-frequency image includes:

[0019] High-resolution micro-Doppler time-frequency images were obtained by suppressing cross-term interference through dual smooth windows in the time and frequency domains. Among them, the micro-Doppler time-frequency images of rotary-wing UAVs showed periodic sinusoidal time-frequency curves, while the micro-Doppler time-frequency images of birds showed composite non-sinusoidal time-frequency curves.

[0020] In one embodiment, constructing the target image fingerprint database based on the target residual neural network includes:

[0021] Micro-Doppler time-frequency images of drones with different parameters and micro-Doppler time-frequency images of different kinds of birds were collected to construct a training sample set;

[0022] Using the target residual neural network as the backbone for feature extraction, the two-dimensional images in the training sample set are mapped into high-dimensional feature vectors to obtain image fingerprints;

[0023] The target image fingerprint database is generated by using a triplet loss optimization network.

[0024] In one embodiment, estimating the UAV state parameters of the low-altitude target to be identified includes:

[0025] The target distance is estimated by performing inverse discrete Fourier transform along the subcarrier direction to obtain the distance of the UAV;

[0026] The target velocity is estimated by performing a discrete Fourier transform along the sign direction to determine the speed of the UAV;

[0027] The number of blades on the UAV rotor is determined by counting the sinusoidal envelope of the micro-Doppler time-frequency image.

[0028] In one embodiment, acquiring the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified includes:

[0029] The echo signal of the low-altitude target to be identified is processed using a preset time-frequency analysis method to obtain the micro-Doppler image; wherein, the preset time-frequency analysis method is a smooth pseudo-Wegener-Vell distribution time-frequency analysis method or a wavelet transform time-frequency analysis method.

[0030] Secondly, this application also provides a low-altitude target detection device, comprising:

[0031] The acquisition module is used to acquire the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified;

[0032] The identification module is used to identify the micro-Doppler image using a target image fingerprint database to obtain the category information of the low-altitude target to be identified; the target image fingerprint database is constructed through a target residual neural network;

[0033] An estimation module is used to estimate the drone state parameters of the low-altitude target to be identified if the category information is a drone, wherein the drone state parameters include the number of blades, distance, and speed.

[0034] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the low-altitude target detection method described in the first aspect.

[0035] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the low-altitude target detection method described in the first aspect.

[0036] In summary, this application proposes a low-altitude target detection method, apparatus, computer device, and readable storage medium, comprising: acquiring a micro-Doppler image corresponding to the echo signal of a low-altitude target to be identified; identifying the micro-Doppler image using a target image fingerprint database to obtain the category information of the low-altitude target to be identified; the target image fingerprint database being constructed through a target residual neural network; and if the category information is an unmanned aerial vehicle (UAV), estimating the UAV state parameters of the low-altitude target to be identified, wherein the UAV state parameters include the number of blades, distance, and speed. This application utilizes residual networks to solve the gradient vanishing problem in deep networks, and by constructing a target image fingerprint database, achieves rapid identification of low-altitude target type information, and simultaneously completes the estimation of UAV state parameters such as distance and speed, supporting integrated low-altitude all-domain perception and improving the accuracy of identifying UAV low-altitude targets in complex environments. Attached Figure Description

[0037] Figure 1 This is an application environment diagram of a low-altitude target detection method in one embodiment;

[0038] Figure 2 This is a geometric diagram of the rotor of a drone in one embodiment;

[0039] Figure 3 A front view of a bird wing model including two joints (elbow joint and wrist joint) in one embodiment;

[0040] Figure 4 This is a top view of a bird wing model that includes two joints (elbow joint and wrist joint) in one embodiment.

[0041] Figure 5 This is a flowchart illustrating a low-altitude target detection method in one embodiment;

[0042] Figure 6 This is a schematic diagram of a target residual neural network model in one embodiment;

[0043] Figure 7 This is a schematic diagram of the residual block in one embodiment;

[0044] Figure 8 This is a schematic diagram of 2D-FFT joint distance-velocity estimation in one embodiment;

[0045] Figure 9 This is a two-dimensional distance-velocity spectrum in one embodiment;

[0046] Figure 10 This is a flowchart illustrating the steps involved in constructing a target image fingerprint database in one embodiment;

[0047] Figure 11This is a time-domain waveform diagram of a rotary-wing drone based on OFDM signals from a 5G base station in one embodiment;

[0048] Figure 12 This is a frequency domain waveform diagram of a rotary-wing drone based on OFDM signals from a 5G base station in one embodiment;

[0049] Figure 13 This is a time-domain waveform diagram of a bird based on a 5G base station OFDM signal in one embodiment;

[0050] Figure 14 This is a frequency domain waveform diagram of a bird based on a 5G base station OFDM signal in one embodiment.

[0051] Figure 15 This is a micro-Doppler time-frequency diagram of a UAV before clutter suppression in one embodiment;

[0052] Figure 16 This is a micro-Doppler time-frequency diagram of a UAV after clutter suppression in one embodiment;

[0053] Figure 17 This is a micro-Doppler time-frequency diagram of a bird before clutter suppression in one embodiment;

[0054] Figure 18 This is a micro-Doppler time-frequency diagram of a bird after clutter suppression in one embodiment;

[0055] Figure 19 This is a structural block diagram of a low-altitude target detection device in one embodiment;

[0056] Figure 20 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] The low-altitude target detection method provided in this application embodiment can be applied to, for example... Figure 1 The application environment is shown. In this embodiment, the low-altitude target detection method is applied to communication base stations located around substations. These communication base stations can be 5G, 6G, or other base stations capable of transmitting Orthogonal Frequency Division Multiplexing (OFDM) signals. It should be noted that the specific type of communication base station in this embodiment can be configured according to the needs of the actual application scenario.

[0059] Specifically, in this embodiment, a low-altitude target refers to a flying target in a low-altitude environment. This flying target can be a rotary-wing drone or a bird. OFDM is a multi-carrier modulation technique whose core principle is to decompose a high-speed data stream into multiple low-speed sub-data streams, which are then transmitted in parallel through a set of orthogonal subcarriers. The orthogonality between subcarriers (spectral overlap without interference) improves spectral efficiency, while a cyclic prefix (CP) is used to resist multipath fading.

[0060] In this embodiment, a composite echo signal model of base station transmitted signal and strong ground clutter is constructed based on the underlying signal system of OFDM-based Integrated Sensing and Communication (ISAC).

[0061] Specifically, the complete baseband transmitted signal model at the transmitter in continuous time. It can be described as:

[0062]

[0063] in, The number of consecutive OFDM symbols in a frame of probe signal. The number of subcarriers, For allocation in the The OFDM symbol, the first Complex modulation symbols on each subcarrier The total physical duration of a single OFDM symbol. , For the duration of the effective symbol, , The frequency spacing between subcarriers, The duration for which the cyclic prefix is ​​inserted before each symbol. Specifically, the total available bandwidth of the system is... The transmitted signal is organized in frames and is evenly divided into There are subcarriers, and the frequency interval between the subcarriers is denoted as . Cyclic prefixes are used to suppress delay spread caused by multipath propagation.

[0064] Due to clutter interference in complex environments, when transmitting signals Radiation extends into low-altitude airspace and via [the following]: static ground clutter scatterers and The continuous-time echo signal vector captured by the base station receiving array after the complex spatial channel reflection of a moving target (a drone or bird with main translational and local micro-motion characteristics) This manifests as the impulse response of the transmitted signal and the spatial multipath channel. Linear convolution:

[0065]

[0066] in, The noise is additive white Gaussian noise. The target channel response is superimposed with high-frequency transient Doppler components caused by micro-motions. The corresponding spatial multipath channel impulse response. for:

[0067]

[0068] in, Represents static path components. Represents the dynamic scattering path component. For path complex gain, For Doppler frequency shift, This is the response vector of the receiving antenna array. To transmit the antenna array response vector, and Used to describe the antenna at an angle radiation patterns For delay The Dirac function, For transmitting filters, This represents the complex gain of the static scatterer. For dynamic scatterer complex gain, and The Doppler frequency shift is caused by the motion of the scattering body. For the first The number of dynamic paths for each scattering cluster The angle of arrival of the scatterer. The scattering path component represents the distance from the scattering object. Specifically, the static path component is used to represent the contribution of the direct path or fixed reflection path of the base station radar line-of-sight (LOS). The dynamic scattering path component is used to represent the multipath component caused by dynamic scattering objects.

[0069] In one embodiment, the low-altitude target detection method further includes:

[0070] Based on the OFDM-based integrated communication and sensing signal mechanism, the micro-Doppler signals of drones and birds are modeled to facilitate the subsequent differentiation of drones and birds based on their micro-Doppler time-frequency characteristics.

[0071] In this embodiment, Integrated Communication and Sensing (ISAC) refers to the fusion of communication and sensing functions into a single system. By sharing hardware, spectrum, signal processing, and other resources, it achieves multiple uses from a single transmission, effectively improving the overall system efficiency and resource utilization. Micro-Doppler refers to the Doppler frequency modulation phenomenon generated by the minute movements of a target or its components on radar echoes. This modulation creates unique sidelobes or frequency variation characteristics in the radar echo spectrum, which can be used to identify the fine motion state of the target.

[0072] refer to Figure 2 This paper provides a geometric diagram of a UAV rotor. For multi-rotor UAVs, the high-speed rotating rigid blades are their most prominent physical structural feature. Specifically, the steps for modeling the micro-Doppler signal of the UAV include:

[0073] The rotor system of a multi-rotor drone consists of: A uniformly distributed leaf, the physical length of a single leaf is denoted as . Assuming the rotor maintains a constant angular velocity. The physical rotation frequency can be expressed as: In electromagnetic scattering theory, a single blade can be considered equivalent to a linear array composed of countless continuous local scattering points. For a blade distributed at the... On each blade, and with a radial length from the center of rotation of... The instantaneous micro-motion projection displacement of any tiny scattering point in the line-of-sight (LOS) direction of the base station radar. It can be approximately derived as follows:

[0074]

[0075] in, This represents the phase angle of the reference blade at the initial moment of observation. The target's spatial elevation angle relative to the base station.

[0076] Based on the Doppler effect principle, the above instantaneous displacement equation is interpreted with respect to time. By solving for the first derivative and substituting it into the traditional Doppler frequency shift formula, the instantaneous micro-Doppler frequency induced by the specific scattering point can be extracted. :

[0077]

[0078] in, This refers to the length of a single blade, or the radial distance of the scattering point. Let be the angular velocity of the rotor. To reference the phase angle of the blade at the initial moment of observation, For the number of leaves, Indicates the leaf index. The target's spatial elevation angle relative to the base station.

[0079] refer to Figure 3 A front view of a bird wing model including two joints (elbow and wrist joints) is provided for reference. Figure 4 This provides a top-down view of a bird's wing model, including two joints (elbow and wrist). It's important to note that the flapping mechanism of birds differs from the rigid body rotation of rotorcraft; their wings incorporate skeletal joints and possess the flexible characteristics of feathers, and their flight relies on the coordinated joint movements between the upper arm and forearm. Specifically, the steps for modeling the micro-Doppler signals of birds include:

[0080] Set the main body of the bird to have a radial velocity Maintain constant speed and level flight, and set the center of its body as the origin of the local coordinate system. Assume the fundamental frequency of flapping wings is... Then the corresponding angular frequency is At the same time, we approximate the single wing of a bird in the wingspan direction as a wing containing... A linear array of tiny scattering points.

[0081] When a bird flaps its wings, the micro-displacement produced by the wings is mainly reflected on the vertical axis perpendicular to the flight surface. The basic amplitude of the wing root near the torso is defined as... The basic amplitude of the outer wingtip at the end is And assume that there is a difference between lateral and medial movement. The phase delay. For the first [element / phase] in the linear array. The instantaneous micro-displacement in the vertical direction at each scattering point. This can be represented as a weighted sum of the medial basic movement and the lateral delayed movement:

[0082]

[0083] in, Represents the initial phase state; and These represent the weighting coefficients of the inner and outer motion components that affect the scattering point, respectively.

[0084] Therefore, the first Instantaneous radial micro-displacement of each scattering point along the line of sight of the base station It can be calculated as:

[0085]

[0086] By analyzing the aforementioned radial displacement with respect to time... By performing differentiation, we can derive the first... Instantaneous micro-motion velocity of each scattering point :

[0087]

[0088] Based on this, and according to the fundamental principles of the Doppler effect, this instantaneous velocity can be converted into the corresponding instantaneous micro-Doppler frequency:

[0089]

[0090] Based on the above steps, the rotor system dynamics model of the corresponding rotary-wing UAV and the flapping wing kinematics model of the corresponding bird can be modeled respectively. The instantaneous micro-Doppler frequency of the UAV... The frequency response curve exhibits sinusoidal modulation characteristics, and its harmonic orders are integer multiples of the number of blades. (Instantaneous micro-Doppler frequency of the bird) The signals exhibit asymmetric modulation characteristics in the time-frequency diagram, with more significant spectral diffusion due to joint coordination and feather flexibility. By modeling the micro-Doppler signals of UAVs and birds using an OFDM-based integrated communication and sensing signal mechanism, it is easier to obtain micro-Doppler images through time-frequency analysis. This allows for the identification of the category information of low-altitude targets based on the micro-Doppler images, thus achieving accurate identification of UAVs and birds.

[0091] In one embodiment, such as Figure 5 As shown, a low-altitude target detection method is provided, which can be applied to, for example... Figure 1 The following steps are used as an example to illustrate the process of establishing a communication base station:

[0092] Step 501: Obtain the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified.

[0093] In this embodiment, the low-altitude target to be identified can refer to a non-cooperative target near the communication base station, or any low-altitude target that the communication base station needs to identify.

[0094] In this embodiment, a preset time-frequency analysis method is used to process the echo signal of the low-altitude target to be identified, thereby obtaining a micro-Doppler image. The preset time-frequency analysis method is either a smoothed pseudo-Wigner-Ville distribution (SPWVD) time-frequency analysis method or a wavelet transform time-frequency analysis method. Specifically, time-frequency analysis is a method that analyzes signals simultaneously in the time and frequency domains to reveal the variation of the signal's frequency components over time. By constructing the energy distribution on the time-frequency plane, time-frequency analysis combines the time and frequency domain information of the signal, making it suitable for analyzing non-stationary signals.

[0095] In this embodiment, the smoothed pseudo-Wigner-Weil distribution, as a quadratic time-frequency analysis method, can accurately depict the micro-Doppler characteristics of low-altitude targets generated by micro-motions such as rotor and wing vibrations. These characteristics are manifested as specific spectral lines or energy distribution patterns on the time-frequency plot. Wavelet transform can analyze signals using basis functions with variable scales (such as Morlet wavelets). Wavelet transform has better time-frequency localization characteristics and is particularly suitable for capturing transient components in micro-Doppler characteristics.

[0096] It should be noted that the actual algorithm of the preset time-frequency analysis method can be determined according to the needs of the actual application scenario. By analyzing the echo signal of the low-altitude target to be identified collected by the communication base station using the preset time-frequency analysis method, a micro-Doppler image containing micro-Doppler features can be obtained. This allows for subsequent comparison of the micro-Doppler image with a target image fingerprint database to obtain the corresponding fingerprint information. It should be noted that the fingerprint information is used to indicate the category of the low-altitude target.

[0097] Step 502: The micro-Doppler image is identified using a target image fingerprint database to obtain the category information of the low-altitude target to be identified. The target image fingerprint database is constructed using a target residual neural network.

[0098] In this embodiment, a preset time-frequency analysis method is used to extract time-frequency images of a large number of drones and birds of different types and parameters. The time-frequency micro-Doppler images with corresponding labels are then fed into a target residual neural network for offline fingerprint database training to obtain a target image fingerprint database. Specifically, the input to the target image fingerprint database is a micro-Doppler image, and the output of the target image fingerprint database is a high-dimensional feature vector. This high-dimensional feature vector corresponds to a unique fingerprint of the input micro-Doppler image and is used to indicate the category information of low-altitude targets.

[0099] In this embodiment, as Figure 6 and Figure 7 As shown, the target residual neural network consists of three parts: an input block, a residual block, and an output block. The residual block is the core module of the target residual neural network model, comprising two main parts: the main path and skip connections. The main path consists of two or three convolutional layers, each followed by a batch normalization layer and an activation function. The goal of the convolutional layers in the main path is to learn a complex representation of the input data. Skip connections directly connect the input to the output of the main path, forming a shortcut. Therefore, the convolutional layers in the main path only need to learn the difference between the input and output, rather than directly learning the complete output features. This embodiment, through the above design of the target residual neural network, effectively helps to alleviate the gradient vanishing problem in backpropagation of deep neural networks, thus enabling the efficient training of extremely deep network structures.

[0100] Specifically, after extracting high-level semantic features through a series of stacked residual blocks, the network proceeds to the output block. The output block typically uses a global average pooling layer to compress the three-dimensional feature tensor into a one-dimensional high-dimensional feature vector. This high-dimensional feature vector is the unique fingerprint of the image.

[0101] In this embodiment, the micro-Doppler image is identified by an offline fingerprint database, which advances the cumbersome model training process. In practical applications, the system can quickly convert the micro-Doppler image to be identified into a unique, high-dimensional feature vector and compare it with the target category features already in the database, thereby achieving rapid classification of low-altitude targets.

[0102] In this embodiment, the category information of low-altitude targets includes drones and birds. If the category information of the low-altitude target to be identified is a bird, no subsequent steps are executed. It should be noted that if the category information of the low-altitude target to be identified is a bird, corresponding deterrence steps can also be initiated according to the needs of the actual application scenario. If the category information of the low-altitude target to be identified is a drone, the drone's state parameters are further identified to provide prior information for subsequently adopting different types of interception methods.

[0103] Step 503: If the category information is UAV, estimate the UAV state parameters of the low-altitude target to be identified, where the UAV state parameters include the number of blades, distance and speed.

[0104] In this embodiment, after high-resolution extraction of the micro-Doppler signal and identification of low-altitude, slow-moving, small-target UAVs using a residual neural network, parameter estimation is further performed on non-cooperative UAVs to achieve reconnaissance. It should be noted that the low-altitude target to be identified in this embodiment is a non-cooperative low-altitude flying target, such as a rotary-wing UAV.

[0105] Specifically, in this embodiment, the target distance can be estimated by performing Inverse Fast Fourier Transform (IFFT) along the subcarrier direction to obtain the distance of the UAV; the target velocity can be estimated by performing Fast Fourier Transform (FFT) along the sign direction to determine the velocity of the UAV; and the number of blades of the UAV rotor can be determined by the number of sinusoidal envelopes in the micro-Doppler time-frequency image.

[0106] Specifically, the number of sine envelopes in the time-frequency plot can be directly observed to determine the number of rotor blades of the UAV. The estimation steps for distance and velocity parameters are as follows:

[0107] Assuming the first receiver in the array is selected The first antenna serves as the observation link. Based on the previously mentioned discretized sensing model, under ideal conditions where noise interference is not considered, the first antenna in the frequency domain... The echo signal of a target mainly includes the following core phase components:

[0108]

[0109] in, This represents the equivalent complex amplitude, which integrates various factors such as the target's scattering characteristics and the array's spatial response. Based on the above equation, it can be seen that the target's subcarrier sequence... and OFDM symbol number These two dimensions exhibit independent complex exponential modulation characteristics. This embodiment extracts velocity and distance parameters by performing joint spectrum estimation on these two dimensions in a two-dimensional discrete domain.

[0110] like Figure 8 As shown, this embodiment simultaneously acquires two-dimensional parameters, for Figure 8 The data matrix is ​​processed by performing inverse discrete Fourier transform in the carrier direction and discrete Fourier transform in the symbol direction, thereby constructing the following two-dimensional joint spatial domain transformation relationship:

[0111]

[0112] Substituting the core phase expression into the above equation and expanding it, we get:

[0113]

[0114] When the two exponential terms in the two-dimensional phase can completely cancel each other out, the global maximum peak can be obtained in the range-Doppler frequency domain spectrum. This requires satisfying the following equation:

[0115]

[0116] This embodiment locks the coordinates of the corresponding peak point by performing a joint optimization operation in the above two-dimensional spectrum diagram. This allows for the estimation of two-dimensional velocity-distance parameters, such as... Figure 9 As shown.

[0117] In one embodiment, such as Figure 10 As shown, before acquiring the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified, the low-altitude target detection method also includes:

[0118] Step 1001: Acquire the composite echo signal of the low-altitude target. The composite echo signal includes static clutter.

[0119] In this embodiment, by means of Figure 1The 5G base station or other types of communication base stations shown acquire composite echo signals from low-altitude targets. Other types of communication base stations include 6G base stations, etc.

[0120] Step 1002: The Pratt circle fitting algorithm is used to perform static clutter suppression on the composite echo signal to obtain the target echo signal after clutter suppression.

[0121] In this embodiment, the static clutter suppression of the composite echo signal using the Pratt circle fitting algorithm includes the following steps:

[0122] First, the single-carrier slow time series of the composite echo signal is extracted and mapped to the IQ complex plane to obtain the original slow time series. Second, a Pratt algebraic circle equation model is constructed, and the generalized eigenvalue problem is solved with constraints to obtain the coordinates of the static clutter center. Third, the clutter component corresponding to the static clutter center coordinates is removed from the original slow time series, and the Doppler rotation component is translated to the origin of the complex plane to complete the clutter suppression of the composite echo signal, thus obtaining the target echo signal.

[0123] Specifically, extract the first... Slow time series on each subcarrier Mapped to the complex plane. The residual static clutter manifests as a fixed DC bias point. The translational echo of the target appears as a standard circular arc trajectory around that point.

[0124] The algebraic circle equation model is constructed using the Pratt circle fitting algorithm:

[0125]

[0126] Define the vector of parameters to be estimated To avoid the problem of divergence or curvature instability in ordinary least squares fitting when the target echo occupies only a small segment of the arc, the Pratt circle fitting algorithm introduces parameter constraints. Represent it in matrix form. Among them, the constraint matrix for:

[0127]

[0128] Extract Substituting OFDM symbol data, an observation matrix can be constructed. :

[0129]

[0130] Based on the least squares criterion and Pratt constraints, the optimal parameter solution can be transformed into the following generalized eigenvalue problem:

[0131]

[0132] By performing eigenvalue decomposition on the above equation, the eigenvector corresponding to the smallest positive eigenvalue is taken as the optimal estimate of the parameter. Therefore, the horizontal and vertical coordinates of the static clutter center can be calculated:

[0133]

[0134] The equivalent center of static clutter is obtained based on the horizontal and vertical coordinates of the static clutter center. Subtract this clutter component from the original slow time series: This allows the Doppler rotation component of low-altitude targets to be translated and reset to the origin of the complex plane, thus suppressing clutter at that angle in the time domain.

[0135] Step 1003: Perform time-frequency analysis on the target echo signal using a smoothed pseudo-Wegener-Vell distribution to obtain a micro-Doppler time-frequency image.

[0136] In this embodiment, a smoothed pseudo-Wigner-Vell distribution is introduced as the core time-frequency analysis tool for training the target residual neural network. In practical applications, micro-Doppler echoes are non-stationary signals whose frequencies fluctuate drastically over time. Traditional global Fourier transforms, lacking time-dimension resolution, cannot accurately characterize the transient evolution of micro-frequency fluctuations. Short-Time Fourier Transforms (STFTs), constrained by the uncertainty principle, present a contradiction between time and frequency resolution. While the Wigner-Vell distribution (WVD) exhibits extremely high time-frequency clustering, its direct use results in severe cross-term interference because drone and bird echoes are typically multi-component signals composed of multiple scattering points.

[0137] Specifically, this embodiment uses dual smooth windows in both the time and frequency domains to suppress cross-term interference and obtain high-resolution micro-Doppler time-frequency images. The micro-Doppler time-frequency images of rotary-wing UAVs exhibit periodic sinusoidal time-frequency curves, while the micro-Doppler time-frequency images of birds exhibit composite non-sinusoidal time-frequency curves.

[0138] This embodiment employs SPWVD by introducing independent smoothing window functions in both the time and frequency domains, effectively suppressing cross-terms while maintaining a time-frequency resolution far exceeding that of STFT. Furthermore, the difference in high-precision time-frequency characteristics can be used to distinguish between drones and birds.

[0139]

[0140] in, Represents the center time of the observation. To analyze frequency, This indicates the complex conjugate operation. It is a frequency domain smoothing window used to reduce cross-term interference in the frequency direction; It is a time-domain smoothing window used to smooth out cross-term interference in the time direction.

[0141] In one embodiment, the actual radar digital signal processing system typically analyzes a sampled discrete signal sequence. Assume the discretized echo sequence is... Discrete time-domain smoothing window is (Window length is) ), frequency domain smoothing window is (Window length is) Then, the corresponding discrete SPWVD calculation formula can be expressed as:

[0142]

[0143] After obtaining a high-resolution time-frequency image of the target without severe cross-term interference through the above transformation, the micro-motion parameters of the target can be extracted and estimated from it.

[0144] Figure 11 and Figure 12 These are the time-domain and frequency-domain waveforms of a rotary-wing drone based on OFDM signals from a 5G base station. Figure 13 and Figure 14 These are the time-domain and frequency-domain waveforms of birds in flight based on OFDM signals from 5G base stations. Figure 15 The UAV micro-Doppler time-frequency map before clutter suppression. Figure 16 The UAV micro-Doppler time-frequency map after clutter suppression, Figure 17 The bird micro-Doppler time-frequency diagram before clutter suppression. Figure 18 The bird micro-Doppler time-frequency diagram after clutter suppression.

[0145] Based on the above steps, such as Figure 11-18 As shown, before clutter suppression, the time-frequency plot exhibits very obvious zero-frequency strong signal interference, making it impossible to observe the micro-motion characteristics of the target. After clutter suppression is achieved through the circle fitting method, it can be clearly observed in the time-frequency plot that the time-frequency curve of the rotor UAV echo signal exhibits a sinusoidal change. This is due to the Doppler frequency change caused by the high-speed circular motion of the scattering point at the rotor tip. Furthermore, the amplitude and frequency of the sine curve are related to the rotational speed and length of the UAV rotor, while the number of envelopes is positively correlated with the number of rotors. The sine curve is distributed around a straight line, where the straight line is the Doppler signal generated by the translational motion of the UAV fuselage, thus allowing the observation of the UAV's translational velocity.

[0146] Bird motion involves complex movements such as flapping and sweeping, therefore, the time-frequency curve is not a simple sine curve. The micro-Doppler characteristics of bird flapping motion become more pronounced with increasing wingspan and flapping frequency. Furthermore, the flapping frequency of birds is much lower than the rotational speed of UAV rotors; therefore, the period and peak micro-Doppler frequencies reflected in the time-frequency image differ significantly from those of UAV rotors, providing a valuable feature basis for deep learning classification based on time-frequency micro-Doppler images.

[0147] Step 1004: Train the target residual neural network using micro-Doppler time-frequency images, and construct the target image fingerprint database based on the target residual neural network.

[0148] In this embodiment, the target residual neural network is trained using the micro-Doppler time-frequency images obtained in steps 1001-1003. It should be noted that in actual application scenarios, the acquisition of the micro-Doppler image of the low-altitude target to be identified in step 501 can also be achieved using steps 1001-1003, so as to improve the resolution and accuracy of the acquired micro-Doppler time-frequency images.

[0149] Specifically, the steps for constructing a target image fingerprint database based on a target residual neural network include:

[0150] First, collect micro-Doppler time-frequency images of drones with different parameters and micro-Doppler time-frequency images of different kinds of birds to construct a training sample set.

[0151] Second, using the target residual neural network as the backbone for feature extraction, the two-dimensional images in the training sample set are mapped into high-dimensional feature vectors to obtain image fingerprints.

[0152] Third, a triplet loss optimization network is used to generate a target image fingerprint database.

[0153] In specific embodiments, the structure of the target residual neural network model can refer to the specific implementation of the foregoing embodiments, and the internal results of the neural network can also be adjusted according to the needs of actual application scenarios, such as adjusting the size of the convolution kernel and the number of residual block layers to adapt to different application scenarios.

[0154] In this embodiment, in order to construct a high-quality fingerprint database, so that fingerprints of similar targets cluster together in the feature space while fingerprints of dissimilar targets are far apart, the triplet loss from metric learning can be used to optimize the network during the training phase:

[0155]

[0156] in, For anchor point images, As a positive sample, For negative samples, This represents the mapping function used by the residual network to extract fingerprints. This represents the square of the Euclidean distance. To control the minimum distance between classes, a margin hyperparameter is used. Through the above structure and training strategy, the system ultimately transforms all extracted images of known birds and rotary-wing UAVs into feature vectors in a high-dimensional space, forming a structured image fingerprint database, providing data support for the rapid retrieval and identification of low-altitude targets.

[0157] In summary, this embodiment provides a low-altitude target detection method that utilizes the micro-Doppler effect caused by the different motion characteristics of UAVs and birds. It combines OFDM signal characteristics to construct an analysis model and a residual neural network image classification method, enabling accurate and efficient identification and classification of UAVs with different rotor parameters and birds of different sizes and species. For example, time-frequency micro-Doppler images can clearly distinguish between the echo signals of rotorcraft UAVs and birds. While quickly distinguishing low-altitude target types, it can also perceive the target's distance and speed, and estimate the UAV rotor parameters using specific algorithms. For distance estimation, the target distance can be accurately calculated by processing the OFDM echo signal. For speed estimation, speed information is obtained through FFT processing of the echo signal rows, providing crucial data support for subsequent targeted interception measures. The low-altitude target detection method provided in this embodiment is applied to communication base stations around substations, effectively solving the problems of difficult and costly low-altitude small target detection in traditional methods. Compared with traditional radar systems that rely on high-power pulsed electromagnetic wave signals, this method leverages the widely deployed 5G base stations, reducing equipment installation and maintenance costs. Compared to vision-based recognition systems, it is not limited by nighttime environments and can simultaneously detect the distance and speed of targets.

[0158] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0159] Based on the same inventive concept, this application also provides a low-altitude target detection device for implementing the low-altitude target detection method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the low-altitude target detection device provided below can be found in the limitations of the low-altitude target detection method described above, and will not be repeated here.

[0160] In one embodiment, such as Figure 19 As shown, a low-altitude target detection device 1900 is provided, including: an acquisition module 1910, an identification module 1920, and an estimation module 1930, wherein:

[0161] The acquisition module 1910 is used to acquire the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified;

[0162] The identification module 1920 is used to identify the micro-Doppler image using a target image fingerprint database to obtain the category information of the low-altitude target to be identified; the target image fingerprint database is constructed through a target residual neural network;

[0163] The estimation module 1930 is used to estimate the drone state parameters of the low-altitude target to be identified if the category information is a drone, wherein the drone state parameters include the number of blades, distance and speed.

[0164] In one embodiment, the low-altitude target detection device further includes:

[0165] The module is specifically used to acquire composite echo signals of low-altitude targets, wherein the composite echo signals include static clutter; the Pratt circle fitting algorithm is used to suppress static clutter in the composite echo signals to obtain clutter-suppressed target echo signals; a smooth pseudo-Wigner-Vell distribution is used to perform time-frequency analysis on the target echo signals to obtain micro-Doppler time-frequency images; the target residual neural network is trained using the micro-Doppler time-frequency images, and the target image fingerprint database is constructed based on the target residual neural network.

[0166] In one embodiment, the construction module is further configured to extract the single-carrier slow time series of the composite echo signal and map the single-carrier slow time series to the IQ complex plane to obtain the original slow time series; construct the Pratt algebraic circle equation model, solve the generalized eigenvalue problem in combination with the constraint conditions, and obtain the coordinates of the static clutter center; remove the clutter component corresponding to the static clutter center coordinates from the original slow time series, and translate the Doppler rotation component to the origin of the complex plane to complete the clutter suppression of the composite echo signal and obtain the target echo signal.

[0167] In one embodiment, the construction module is further configured to suppress cross-term interference through a dual smooth window in the time and frequency domains to obtain a high-resolution micro-Doppler time-frequency image; wherein the micro-Doppler time-frequency image of the rotary-wing UAV presents a periodic sinusoidal time-frequency curve, and the micro-Doppler time-frequency image of the bird presents a composite non-sinusoidal time-frequency curve.

[0168] In one embodiment, the construction module is further configured to collect micro-Doppler time-frequency images of drones with different parameters and micro-Doppler time-frequency images of different types of birds to construct a training sample set; using the target residual neural network as the feature extraction backbone, the two-dimensional images in the training sample set are mapped into high-dimensional feature vectors to obtain image fingerprints; and a triplet loss optimization network is used to generate the target image fingerprint database.

[0169] In one embodiment, the estimation module 1930 is further configured to perform inverse discrete Fourier transform processing along the subcarrier direction to estimate the target distance and obtain the distance of the UAV; perform discrete Fourier transform processing along the sign direction to estimate the target velocity and determine the velocity of the UAV; and determine the number of blades of the UAV rotor by the number of sinusoidal envelopes of the micro-Doppler time-frequency image.

[0170] In one embodiment, the acquisition module 1910 further processes the echo signal of the low-altitude target to be identified using a preset time-frequency analysis method to obtain the micro-Doppler image; wherein the preset time-frequency analysis method is a short-time Fourier transform time-frequency analysis method or a wavelet transform time-frequency analysis method.

[0171] In summary, this embodiment provides a low-altitude target detection device that utilizes the micro-Doppler effect caused by the different motion characteristics of UAVs and birds. It combines OFDM signal characteristics to construct an analysis model and a residual neural network image classification method, enabling accurate and efficient identification and classification of UAVs with different rotor parameters and birds of different sizes and species. For example, time-frequency micro-Doppler images can clearly distinguish between the echo signals of rotorcraft UAVs and birds. While quickly distinguishing low-altitude target types, it can also perceive the distance and speed of targets and estimate UAV rotor parameters through specific algorithms. For distance estimation, the target distance can be accurately calculated by processing the OFDM echo signal. For speed estimation, speed information is obtained through FFT processing of the echo signal rows, providing crucial data support for subsequent targeted interception measures. The low-altitude target detection device provided in this embodiment is deployed within communication base stations around substations, effectively solving the problems of difficult and costly low-altitude small target detection in traditional methods. Compared with traditional radar systems that rely on high-power pulsed electromagnetic wave signals, this method utilizes widely deployed 5G base stations, reducing equipment installation and maintenance costs. Compared to vision-based recognition systems, it is not limited by nighttime environments and can simultaneously detect the distance and speed of targets.

[0172] Each module in the aforementioned low-altitude target detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0173] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 20 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a low-altitude target detection method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0174] Those skilled in the art will understand that Figure 20 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0175] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0176] Acquire the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified;

[0177] The micro-Doppler image is identified using a target image fingerprint database to obtain the category information of the low-altitude target to be identified; the target image fingerprint database is constructed through a target residual neural network.

[0178] If the category information is a drone, estimate the drone state parameters of the low-altitude target to be identified, wherein the drone state parameters include the number of blades, distance, and speed.

[0179] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0180] Acquire the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified;

[0181] The micro-Doppler image is identified using a target image fingerprint database to obtain the category information of the low-altitude target to be identified; the target image fingerprint database is constructed through a target residual neural network.

[0182] If the category information is a drone, estimate the drone state parameters of the low-altitude target to be identified, wherein the drone state parameters include the number of blades, distance, and speed.

[0183] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0184] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0185] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for detecting low-altitude targets, characterized in that, include: Acquire the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified; The micro-Doppler image is identified using a target image fingerprint database to obtain the category information of the low-altitude target to be identified; The target image fingerprint database is constructed using a target residual neural network; If the category information is a drone, estimate the drone state parameters of the low-altitude target to be identified, wherein the drone state parameters include the number of blades, distance, and speed.

2. The method according to claim 1, characterized in that, The method further includes: Acquire composite echo signals of low-altitude targets, wherein the composite echo signals include static clutter; The Pratt circle fitting algorithm is used to perform static clutter suppression on the composite echo signal to obtain the target echo signal after clutter suppression. A smooth pseudo-Wegener-Weil distribution was used to perform time-frequency analysis on the target echo signal to obtain a micro-Doppler time-frequency image; The target residual neural network is trained using the micro-Doppler time-frequency image, and the target image fingerprint database is constructed based on the target residual neural network.

3. The method according to claim 2, characterized in that, The step of using the Pratt circle fitting algorithm to perform static clutter suppression on the composite echo signal to obtain the clutter-suppressed target echo signal includes: Extract the single-carrier slow time sequence of the composite echo signal and map the single-carrier slow time sequence to the IQ complex plane to obtain the original slow time sequence; A Pratt algebraic circle equation model is constructed, and the generalized eigenvalue problem is solved by combining the constraints to obtain the coordinates of the static clutter center. The clutter component corresponding to the static clutter center coordinate is removed from the original slow time series, and the Doppler rotation component is translated to the origin of the complex plane to complete the clutter suppression of the composite echo signal, thereby obtaining the target echo signal.

4. The method according to claim 2, characterized in that, The step of performing time-frequency analysis on the target echo signal using a smooth pseudo-Wigner-Weil distribution to obtain a micro-Doppler time-frequency image includes: High-resolution micro-Doppler time-frequency images were obtained by suppressing cross-term interference through dual smooth windows in the time and frequency domains. Among them, the micro-Doppler time-frequency images of rotary-wing UAVs showed periodic sinusoidal time-frequency curves, while the micro-Doppler time-frequency images of birds showed composite non-sinusoidal time-frequency curves.

5. The method according to claim 2, characterized in that, The construction of the target image fingerprint database based on the target residual neural network includes: Micro-Doppler time-frequency images of drones with different parameters and micro-Doppler time-frequency images of different kinds of birds were collected to construct a training sample set; Using the target residual neural network as the backbone for feature extraction, the two-dimensional images in the training sample set are mapped into high-dimensional feature vectors to obtain image fingerprints; The target image fingerprint database is generated by using a triplet loss optimization network.

6. The method according to claim 1, characterized in that, The estimation of the UAV state parameters of the low-altitude target to be identified includes: The target distance is estimated by performing inverse discrete Fourier transform along the subcarrier direction to obtain the distance of the UAV; The target velocity is estimated by performing a discrete Fourier transform along the sign direction to determine the speed of the UAV; The number of blades on the UAV rotor is determined by counting the sinusoidal envelope of the micro-Doppler time-frequency image.

7. The method according to claim 1, characterized in that, The process of acquiring the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified includes: The echo signal of the low-altitude target to be identified is processed using a preset time-frequency analysis method to obtain the micro-Doppler image; wherein, the preset time-frequency analysis method is a smoothed pseudo-Wigner-Vell distribution analysis method or a wavelet transform time-frequency analysis method.

8. A low-altitude target detection device, characterized in that, The device includes: The acquisition module is used to acquire the micro-Doppler image corresponding to the echo signal of the low-altitude target to be identified; The identification module is used to identify the micro-Doppler image using a target image fingerprint database to obtain the category information of the low-altitude target to be identified; the target image fingerprint database is constructed through a target residual neural network; An estimation module is used to estimate the drone state parameters of the low-altitude target to be identified if the category information is a drone, wherein the drone state parameters include the number of blades, distance, and speed.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the low-altitude target detection method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the low-altitude target detection method according to any one of claims 1 to 6.

Images