Unmanned aerial vehicle micro-Doppler identification method and system based on body micro-motion decoupling and double-flow collaborative awareness
By using a volumetric micro-motion decoupling and dual-stream collaborative sensing method, the strong fuselage signal and weak rotor signal in the radar echo are effectively decoupled. A main path and auxiliary path neural network are constructed to extract and fuse macroscopic and microscopic features, solving the misjudgment problem in UAV identification and achieving higher accuracy and robustness in identification.
Patent Information
- Application Number
- CN202610169012.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-05
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies struggle to effectively decouple strong fuselage signals from weak rotor signals in radar echoes, resulting in the masking of micro-Doppler features. Furthermore, deep learning networks tend to ignore key micro-motion details, leading to misjudgments in UAV identification and limited generalization capabilities.
A method based on volumetric micro-motion decoupling and dual-flow collaborative sensing is adopted. The original time-frequency map is generated by short-time Fourier transform, and volumetric motion component suppression and micro-motion enhancement are performed. Main and auxiliary branch neural networks are constructed to extract macroscopic and microscopic features respectively, and feature fusion recognition is performed.
It significantly improves the accuracy and environmental adaptability of UAV recognition, ensures the full learning and utilization of micro-motion features, reduces the probability of misjudgment and missed judgment, and provides more comprehensive feature representation and reliable recognition results.
Smart Images
Figure CN122063555A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of radar signal processing and intelligent recognition technology, and relates to a micro-Doppler recognition method and system for unmanned aerial vehicles based on volume micro-motion decoupling and dual-stream collaborative sensing. Background Technology
[0002] With the booming development of the low-altitude economy, consumer drones are increasingly used in logistics delivery, inspection and photography, and other fields. However, this also brings increasingly serious airspace security challenges such as illegal intrusion and privacy theft. In counter-drone surveillance systems, radar has become a core sensor due to its all-weather and long-range detection advantages. Because the low-altitude environment is extremely complex, drone targets often exhibit low-altitude, small-size, and slow-moving characteristics, and their trajectories are easily confused with biological targets such as birds. The micro-Doppler effect, as the acoustic fingerprint of a target, can reflect the microscopic dynamic characteristics of drone rotor rotation or bird wing flapping, and is a key physical basis for distinguishing between the two.
[0003] However, in real-world radar detection scenarios, target recognition technology based on micro-Doppler features faces severe challenges. First, the echo signal of a UAV is essentially a superimposed signal with extremely uneven energy, composed of a strong body motion component generated by fuselage translation and a weak micro-motion component generated by rotor rotation. Since the radar cross section (RCS) of the fuselage is typically much larger than that of the rotor blades, the high-intensity fuselage signal acts like a searchlight in time-frequency analysis, easily masking weak rotor features through sidelobe leakage or main lobe broadening. Especially at long distances or in low signal-to-noise ratio environments, crucial micro-motion scintillation features are often obscured by noise or strong clutter, making subsequent feature extraction extremely difficult.
[0004] Furthermore, existing intelligent recognition methods often directly employ convolutional neural networks (CNNs) for end-to-end feature learning from the original radar time-frequency images. However, deep learning networks exhibit an inherent inertia, tending to capture salient features. When the input image contains both high-energy, low-frequency fuselage trajectory features and low-energy, high-frequency rotor texture features, the network often prioritizes converging on salient fuselage features to quickly reduce the loss function, thus ignoring the micro-motion details crucial for classification. This feature learning bias makes the model highly susceptible to misclassification when faced with targets with similar fuselage motion states due to a lack of microscopic discriminative information, severely limiting its generalization ability.
[0005] While existing technologies attempt to suppress low-frequency signals using simple frequency domain filtering or windowing, these methods are typically based on fixed prior thresholds and lack adaptability to the time-varying characteristics of signals. They are prone to disrupting the phase continuity of micro-motion signals or causing spectral aliasing while filtering out fuselage signals. Therefore, effectively decoupling strong and weak signals at the physical signal processing level and constructing a collaborative sensing architecture that can simultaneously consider macroscopic motion trends and microscopic rotational details has become a key technical challenge for improving the recognition performance of low-altitude unmanned aerial vehicles (UAVs). Summary of the Invention
[0006] In view of this, the purpose of this invention is to provide a micro-Doppler identification method and system for unmanned aerial vehicles based on volume micro-motion decoupling and dual-stream collaborative sensing.
[0007] To achieve the above objectives, the present invention provides the following technical solution: A micro-Doppler identification method for unmanned aerial vehicles (UAVs) based on volume micro-motion decoupling and dual-stream collaborative sensing includes the following steps: S1: Acquire the echo signal of low-altitude targets detected by radar, generate the original time-frequency map through short-time fourier transform (STFT), and use it as the input of the main branch; S2: Based on the energy difference and spectral distribution characteristics of body motion and micro-motion signals, the original signal is preprocessed to suppress body motion components, and a micro-motion enhancement time-frequency map is generated as the input of the auxiliary branch. S3: Construct a dual-stream collaborative sensing neural network. The main branch network extracts macroscopic motion trend features from the original time-frequency map, and the auxiliary branch network extracts microscopic rotor scintillation features from the micro-motion enhanced time-frequency map. S4: Perform channel splicing and fusion of the main road and auxiliary road features to generate a joint feature vector; S5: Input the joint feature vector into the classification head to complete the identification of UAV targets.
[0008] Furthermore, in S1, the physical model construction of the low-altitude UAV echo signal s(t) based on the radar line-of-sight direction follows the superposition principle:
[0009] Where n(t) is additive white Gaussian noise; The strong scattering component generated by the fuselage translation is mathematically expressed as follows:
[0010] The micro-Doppler modulation component generated by the rotation of the rotor components is mathematically expressed as follows:
[0011] In the formula, and These are the radar cross section (RCS) of the main fuselage and the k-th rotor blade, respectively. The initial distance, v Radial velocity, For radar wavelength, For the blade length, The angular velocity of the rotor. This is the initial phase.
[0012] Furthermore, the specific calculation process for generating the original time-frequency diagram in S1 is as follows: Apply a window function w[n] to the discretized echo signal s[n] and perform a sliding discrete Fourier transform:
[0013] Where m is the time frame index, k is the frequency index, N is the window length, and R is the sliding step size; Take its magnitude to generate the original time-frequency diagram matrix. :
[0014] Furthermore, step S2 includes the following sub-steps: S21: Construct the time-domain fuselage reference vector and calculate the statistical mean vector of the echo signal s[m,n] along the slow time dimension. :
[0015] Where m is the slow-time pulse index, n is the fast-time sampling point index, and M is the total number of pulses; S22: Perform coherent background cancellation and extract the residual signal. :
[0016] S23: Enhanced nonlinearity of the micro-motion signal, generating an enhanced time-domain signal. :
[0017]
[0018] in, To prevent the loss of a tiny amount, The contrast stretching factor is ∠, which indicates the phase. S24: Enhanced signal Perform a short-time Fourier transform (STFT) to generate a micro-motion enhanced time-frequency diagram. .
[0019] Furthermore, in the dual-stream collaborative sensing neural network constructed in S3: The input to the main branch network is the original time-frequency diagram. It contains L convolutional layers for extracting strong signal features from the fuselage. ; The input to the auxiliary branch network is a micro-motion enhanced time-frequency diagram. The network structure is the same as the main path but does not share weights, and it is used to extract weak rotor signal features. ; The calculation process of the convolutional layer is described as follows:
[0020] in This represents the convolution operation. is the activation function for the Rectified Linear Unit (ReLU), where W and b are learnable parameters.
[0021] Furthermore, the specific feature fusion and classification process for S4 and S5 is as follows: S41: Characteristics of the main road and auxiliary road characteristics Perform Global Average Pooling (GAP) on each feature vector to stretch it into a one-dimensional feature vector. and ; S42: Perform feature channel splicing:
[0022] S43: The fused features are mapped to the class space through a fully connected layer (FC), and the probability of belonging to the i-th class of objects is calculated. :
[0023] in C The total number of target categories, , and These are the learnable parameters for the fully connected layer.
[0024] A micro-Doppler identification system for unmanned aerial vehicles (UAVs) based on volume micro-motion decoupling and dual-stream collaborative sensing includes a data acquisition layer and an intelligent processing layer. The data acquisition layer includes a radar transceiver front-end, an analog-to-digital conversion module, and a preprocessing module. The radar transceiver front-end is used to transmit continuous waves or pulse Doppler waveforms and receive target echoes. The analog-to-digital conversion module realizes the conversion of analog signals to digital signals, and the preprocessing module completes signal filtering and noise reduction. The intelligent processing layer includes a main path feature extraction module, a micro-motion enhancement preprocessing module, an auxiliary path feature extraction module, and a dual-stream fusion inference module. The micro-motion enhancement preprocessing module executes body motion suppression and micro-motion enhancement algorithms to generate auxiliary path inputs. The main path feature extraction module and the auxiliary path feature extraction module extract the main path and auxiliary path features respectively. The dual-stream fusion inference module completes feature splicing and classification recognition.
[0025] Furthermore, the micro-motion enhancement preprocessing module is configured to perform the following operations: Construct a time-domain fuselage reference vector and calculate the statistical mean vector of the echo signal s[m,n] along the slow time dimension. Where m is the slow-time pulse index, n is the fast-time sampling point index, and M is the total number of pulses; Perform coherent background cancellation and extract residual signals. ; Nonlinear significance enhancement is performed on the residual signal to generate the enhanced time-domain signal. ; Perform a short-time Fourier transform on the enhanced signal to generate a micro-motion enhanced time-frequency map. .
[0026] The beneficial effects of this invention are as follows: (1) This invention does not rely on neural networks to autonomously learn and separate strong and weak features. Instead, it innovatively introduces a signal enhancement preprocessing step based on coherent background cancellation and nonlinearity at the data input end. This step actively suppresses high-energy fuselage translational components at the physical level and significantly enhances weak rotor micro-motion components, generating an enhanced time-frequency map specifically for micro-feature extraction. This eliminates the need for subsequent deep learning models to painstakingly extract weak features from mixed strong signals, fundamentally avoiding the "laziness" problem of traditional single-stream networks that tend to learn salient features while ignoring key micro-motion details, thus ensuring the full learning and utilization of rotor micro-features.
[0027] (2) This invention designs a collaborative perception neural network with main and auxiliary parallel paths and non-shared weights. The main path branch focuses on extracting the macroscopic motion trend of the target from the original time-frequency map, such as flight speed and trajectory; the auxiliary path branch is forced to focus on the microscopic motion details of the rotor from the enhanced time-frequency map, such as rotational speed, number of blades, and other flashing features. Finally, through feature fusion, macroscopic motion information and microscopic "fingerprint" information are effectively combined to form a more comprehensive and refined feature representation of the UAV target, which significantly improves the inter-class discrimination and recognition accuracy.
[0028] (3) Due to the adoption of an adaptive nonlinear gain mechanism, this invention can enhance the rotor micro-motion signal that was originally buried in noise to an intensity level that can be effectively sensed by the network. This enhancement process enables the system to stably extract key micro-Doppler features even when facing complex low-altitude environments such as long-distance detection, low signal-to-noise ratio, or background clutter, demonstrating excellent environmental adaptability and robustness, and greatly reducing the probability of misjudgment and missed judgment.
[0029] (4) Compared with the "black box" end-to-end single-stream network, the dual-path processing flow of the present invention has stronger physical meaning and interpretability. The main path and the auxiliary path each perform their own functions, and the source of features is clear, so that the final classification decision is based on dual evidence of "macro behavior + micro structure", which enhances the credibility of the identification results and provides a clear direction for system performance analysis and optimization.
[0030] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description
[0031] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 The diagram shows the overall architecture of the UAV micro-Doppler identification method and system based on volume micro-motion decoupling and dual-stream collaborative sensing. Figure 2 Flowchart of the algorithm for enhancing preprocessing micro-motion features; Figure 3 This is a schematic diagram of the structure of a dual-stream collaborative sensing neural network. Detailed Implementation
[0032] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.
[0033] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0034] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.
[0035] like Figure 1 As shown, the method provided by this invention mainly includes four core stages: signal acquisition and modeling, micro-motion feature enhancement, dual-stream feature extraction and fusion classification. The specific implementation steps are as follows: S1: Echo signal modeling and main input generation In this embodiment, the radar transmits a linear frequency modulated continuous wave (LFMCW) or a pulse Doppler waveform. Assume the echo signal of a low-altitude UAV target in the radar's line-of-sight (LOS) direction. Translational component of the fuselage Rotor micro-motion components and noise Its composition and physical model are constructed as follows:
[0036] The strong scattering component generated by the fuselage translation is described as follows:
[0037] The micro-Doppler modulation component generated by the rotation of the rotor components is described as follows:
[0038] In the formula, and They are respectively the main body of the fuselage and the first Radar cross section (RCS) of each rotor blade. The initial distance, Radial velocity, For radar wavelength, For the blade length, ω represents the angular velocity of the rotor.
[0039] Key constraints: Due to limitations imposed by the actual physical structure, (The difference is usually greater than 10dB), causing the original signal to... The amplitude envelope is mainly composed of Decide.
[0040] Discretized echo signal Apply Hamming window Then, a short-time Fourier transform (STFT) is performed to generate the original time-frequency diagram. :
[0041]
[0042] Should It can clearly reflect the Doppler frequency shift of the fuselage and serve as the input for the main branch.
[0043] S2: Significantly enhanced body movement inhibition and micromotion (auxiliary pathway input generation) To address the issue of CNNs "ignoring" weak signals under strong signal conditions, this step generates auxiliary input through time-domain signal processing. For example... Figure 2 As shown, it specifically includes: S21: Construct the time-domain fuselage reference vector. Assume that within a coherent processing interval (CPI), the fuselage translational components are stationary in the slow-time dimension (frequency approximately constant, phase linearly changing), while the rotor signal fluctuates rapidly. Calculate the echo signal matrix. Statistical mean vector along the slow time dimension :
[0044] in, For slow-time pulse indexing, For fast time sampling point index, This represents the total number of pulses.
[0045] S22: Perform coherent background cancellation. Subtract the fuselage reference vector from each pulse echo of the original echo matrix, using the principle of coherent subtraction to eliminate strong fuselage background and extract the residual signal. :
[0046] This step directly strips away the main energy (fuselage echo) in the time domain, while retaining the rotor micro-motion components and noise that modulate rapidly over time.
[0047] S23: Significant enhancement of nonlinearity in the micro-motion signal. To address the issue that the rotor echo energy in the residual signal remains weak (potentially close to the noise floor), this embodiment introduces an adaptive gain control factor. With logarithmic mapping mechanism.
[0048] First, calculate the adaptive gain factor. This is used to measure the degree of attenuation of a micro-motion signal relative to the original strong signal.
[0049] Subsequently, a nonlinear enhancement function is constructed to generate the enhanced time-domain signal. :
[0050] in, To prevent the loss of a tiny amount, This is the contrast stretching factor (empirical value is 10-100). This indicates phase taking. The function of this formula is to increase the intensity of the originally weak rotor signal to the same order of magnitude as the original fuselage signal, while using a logarithmic function to suppress high-amplitude burst impulse noise.
[0051] S24: Time-frequency domain mapping. For the enhanced signal... Perform the same STFT transformation as S1 to generate a micro-motion enhanced time-frequency diagram. At this point, the zero-frequency strong signal in the diagram has been removed, and the rotor scintillation bands on both sides are clearly visible, which are used as the input for the auxiliary branch.
[0052] S3: Constructing a dual-stream collaborative sensing neural network like Figure 3 As shown, the network contains two parallel branches: 1. Main Path Branch (Body Stream): Input is The original time-frequency diagram A lightweight ResNet-18 backbone network is used to extract low-frequency, high-energy trajectory features. .
[0053] 2. Auxiliary Path Branch (Micro-Motion Stream): The input is a micro-motion enhanced time-frequency plot of the same size. Also employing the ResNet-18 structure (but without sharing weights), this branch is forced to focus on learning high-frequency, periodic rotor micro-motion texture features because fuselage interference has been removed from the input image. .
[0054] The mathematical operations of a convolutional layer are described as follows:
[0055] in The convolution operation is represented. This is the ReLU activation function.
[0056] S4 and S5: Feature Fusion and Refined Recognition Before the classification head, global average pooling (GAP) is performed on the feature maps of the two branches to obtain two one-dimensional feature vectors. and .
[0057] Perform feature channel splicing:
[0058] concatenated feature vectors It integrates complementary information from macroscopic motion and microscopic structure.
[0059] Finally, the fused features are mapped to the class space through a fully connected layer (FC), and the classification result is output using the Softmax function.
[0060] in For the target to belong to the first The probability of a class (such as a quadcopter drone, a single-rotor helicopter, a bird, etc.).
[0061] The network training uses the cross-entropy loss function and updates the network parameters of the main path and auxiliary path simultaneously through the backpropagation algorithm.
[0062] Through the above steps, the present invention effectively overcomes the feature selection bias of a single input network when processing radar echoes with vastly different energy levels, and achieves high-precision identification of low-altitude UAVs.
[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A micro-Doppler identification method for unmanned aerial vehicles (UAVs) based on volume micro-motion decoupling and dual-stream collaborative sensing, characterized in that: Includes the following steps: S1: Acquire the echo signal of low-altitude targets detected by radar, generate the original time-frequency diagram through short-time Fourier transform (STFT), and use it as the input of the main branch; S2: Based on the energy difference and spectral distribution characteristics of body motion and micro-motion signals, the original signal is preprocessed to suppress body motion components, and a micro-motion enhancement time-frequency map is generated as the input of the auxiliary branch. S3: Construct a dual-stream collaborative sensing neural network. The main branch network extracts macroscopic motion trend features from the original time-frequency map, and the auxiliary branch network extracts microscopic rotor scintillation features from the micro-motion enhanced time-frequency map. S4: Perform channel splicing and fusion of the main road and auxiliary road features to generate a joint feature vector; S5: Input the joint feature vector into the classification head to complete the identification of UAV targets.
2. The UAV micro-Doppler identification method based on volume micro-motion decoupling and dual-stream collaborative sensing according to claim 1, characterized in that: In S1, the physical model construction of the low-altitude UAV echo signal s(t) based on the radar line-of-sight direction follows the superposition principle: Where n(t) is additive white Gaussian noise; The strong scattering component generated by the fuselage translation is mathematically expressed as follows: The micro-Doppler modulation component generated by the rotation of the rotor components is mathematically expressed as follows: In the formula, and These are the radar cross-sections (RCS) of the main fuselage and the kth rotor blade, respectively. The initial distance, v Radial velocity, For radar wavelength, For the blade length, The angular velocity of the rotor. This is the initial phase.
3. The UAV micro-Doppler identification method based on volume micro-motion decoupling and dual-stream cooperative sensing according to claim 1, characterized in that: The specific calculation process for generating the original time-frequency diagram in S1 is as follows: Apply a window function w[n] to the discretized echo signal s[n] and perform a sliding discrete Fourier transform: Where m is the time frame index, k is the frequency index, N is the window length, and R is the sliding step size; Take its magnitude to generate the original time-frequency diagram matrix. : 。 4. The UAV micro-Doppler identification method based on volume micro-motion decoupling and dual-stream cooperative sensing according to claim 1, characterized in that: S2 includes the following sub-steps: S21: Construct the time-domain fuselage reference vector and calculate the statistical mean vector of the echo signal s[m,n] along the slow time dimension. : Where m is the slow-time pulse index, n is the fast-time sampling point index, and M is the total number of pulses; S22: Perform coherent background cancellation and extract the residual signal. : S23: Enhanced nonlinearity of the micro-motion signal, generating an enhanced time-domain signal. : in, To prevent the loss of a tiny amount, The contrast stretching factor is ∠, which indicates the phase. S24: Enhanced signal Perform a short-time Fourier transform (STFT) to generate a micro-motion enhanced time-frequency diagram. .
5. The UAV micro-Doppler identification method based on volume micro-motion decoupling and dual-stream collaborative sensing according to claim 1, characterized in that: In the dual-stream collaborative sensing neural network constructed in S3: The input to the main branch network is the original time-frequency diagram. It contains L convolutional layers for extracting strong signal features from the fuselage. ; The input to the auxiliary branch network is a micro-motion enhanced time-frequency diagram. The network structure is the same as the main path but does not share weights, and it is used to extract weak rotor signal features. ; The calculation process of the convolutional layer is described as follows: in This represents the convolution operation. is the activation function for the Rectified Linear Unit (ReLU), where W and b are learnable parameters.
6. The UAV micro-Doppler identification method based on volume micro-motion decoupling and dual-stream cooperative sensing according to claim 1, characterized in that: The specific feature fusion and classification process for S4 and S5 is as follows: S41: Characteristics of the main road and auxiliary road characteristics Perform global average pooling (GAP) separately to stretch the vectors into one-dimensional feature vectors. and ; S42: Perform feature channel splicing: S43: The fused features are mapped to the class space through a fully connected layer (FC), and the probability of belonging to the i-th class of objects is calculated. : in C The total number of target categories, , and These are the learnable parameters for the fully connected layer.
7. A micro-Doppler identification system for unmanned aerial vehicles (UAVs) based on volume micro-motion decoupling and dual-stream collaborative sensing, characterized in that: It includes a data acquisition layer and an intelligent processing layer; The data acquisition layer includes a radar transceiver front-end, an analog-to-digital conversion module, and a preprocessing module. The radar transceiver front-end is used to transmit continuous waves or pulse Doppler waveforms and receive target echoes. The analog-to-digital conversion module realizes the conversion of analog signals to digital signals, and the preprocessing module completes signal filtering and noise reduction. The intelligent processing layer includes a main path feature extraction module, a micro-motion enhancement preprocessing module, an auxiliary path feature extraction module, and a dual-stream fusion inference module. The micro-motion enhancement preprocessing module executes body motion suppression and micro-motion enhancement algorithms to generate auxiliary path inputs. The main path feature extraction module and the auxiliary path feature extraction module extract the main path and auxiliary path features respectively. The dual-stream fusion inference module completes feature splicing and classification recognition.
8. The UAV micro-Doppler identification system based on volume micro-motion decoupling and dual-stream cooperative sensing according to claim 7, characterized in that: The micro-motion enhancement preprocessing module is configured to perform the following operations: Construct a time-domain fuselage reference vector and calculate the statistical mean vector of the echo signal s[m,n] along the slow time dimension. Where m is the slow-time pulse index, n is the fast-time sampling point index, and M is the total number of pulses; Perform coherent background cancellation and extract residual signals. ; Nonlinear significance enhancement is performed on the residual signal to generate the enhanced time-domain signal. ; Perform a short-time Fourier transform on the enhanced signal to generate a micro-motion enhanced time-frequency map. .