Target object radar micro-motion signal simulation method and target object intelligent detection method

Abstract

This invention discloses a method for simulating radar micro-motion signals of target objects and an intelligent detection method for target objects, including: constructing a parameterized model library containing the target object, the surrounding environment, and radar sensors; generating a scene configuration file by sampling parameters; using a hybrid algorithm combining the bouncing ray method and physical optics method, combined with the target object motion model, to dynamically calculate the radar echo and generate the original baseband signal; and converting the signal into a time-spectrum graph after signal processing. The scene configuration file during simulation is parsed to automatically generate labels from the time-spectrum graph, constructing a dataset; a deep learning model is trained to infer from the measured time-spectrum graph and output the target object detection result. This invention generates massive amounts of finely labeled radar micro-motion data through high-fidelity physical simulation, effectively solving the problems of difficulty and high cost in acquiring real data, and significantly improving the accuracy and robustness of non-contact target object detection in scenarios such as logistics security inspection.

Description

Technical Field

[0001] This invention relates to the technical field of radar detection technology, computational electromagnetics simulation, artificial intelligence and signal processing integration, and more specifically to a method for intelligent detection of target objects in logistics security inspection scenarios by generating radar micro-motion signals of target objects through high-fidelity physical simulation and using deep learning. Background Technology

[0002] Current mainstream technologies in logistics security inspection mainly include X-ray imaging, manual inspection, and metal detection. While these technologies are highly effective at detecting high-density objects like metal, they have significant limitations in detecting exotic pets. First, their accuracy is insufficient. X-ray imaging relies primarily on differences in material density; exotic pets have similar densities to clothing, foam, and other fillers in packages, making it difficult to create clear and distinguishable image features, leading to frequent missed detections and misjudgments. Manual inspection, on the other hand, depends on the experience of security personnel, is highly subjective, and inefficient, failing to meet the demands of rapid security checks on massive volumes of logistics packages. Second, they lack the ability to differentiate between exotic pets and non-target objects such as pet specimens and toys. Furthermore, some security inspection technologies require opening the package for inspection, which not only damages the integrity of the package but also risks the loss or damage of items inside, while increasing the risk of security personnel coming into contact with unknown pathogens.

[0003] Radar micro-motion detection technology achieves non-contact identification by capturing the frequency modulation characteristics of radar echoes caused by minute movements such as heartbeat, breathing, and limb twitching of target objects. However, the application of this technology in logistics security inspection scenarios faces two major bottlenecks. First, high-quality training data is scarce. Obtaining radar data on exotic pets in logistics parcel scenarios requires building diverse parcel environments and simulating the activity states of different types of exotic pets. This is not only costly and complex to operate, but also poses animal ethics and biosafety risks, making it difficult to form large-scale, multi-scenario labeled datasets. Second, traditional simulation methods have low fidelity. Existing radar signal simulations are mostly aimed at rigid targets or simple moving targets, without considering the complex environmental characteristics of logistics parcels and the physiological rhythms of exotic pets. The generated simulation signals deviate significantly from the real scene, and the performance of detection models trained on such data drops sharply in practical applications.

[0004] Therefore, there is an urgent need for a method to efficiently generate high-fidelity radar micro-motion signals of target objects in logistics security inspection scenarios, as well as an intelligent detection method for target objects, which is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a target object radar micro-motion signal simulation method and a target object intelligent detection method to overcome or at least partially solve the above problems, provide data support for detection algorithm training, and thus realize accurate, fast and non-contact detection of exotic pets in logistics packages.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a method for simulating radar micro-motion signals of a target object, comprising the following steps: S1: Construct a parametric model library; the parameter space of the parametric model library includes at least a target object model library, a packaging environment model library, and a radar sensor model library; wherein, the target object model library is used to store adjustable parameters of the kinematic characteristics and electromagnetic scattering characteristics of the target object; the packaging environment model library is used to store the random scattering field of the packaging environment in which the target object is located, and the electromagnetic parameters of the packaging material; the radar sensor model library is used to store the operating parameters of the radar system; S2: Based on the parameterized model library, sample and combine the parameter space to generate multiple structured scene configuration files describing different virtual simulation scenarios; S3: For each scenario configuration file, a hybrid electromagnetic simulation algorithm is used to dynamically calculate the radar echo of the target object at each moment, and generate the original radar baseband signal containing the micro-Doppler characteristics of the target object. S4: Perform signal processing on the original radar baseband signal to generate a time-spectrum diagram for visually displaying the micro-motion characteristics of the target.

[0008] Preferably, the construction of the target object model library in S1 includes: The kinematic features include storing the motion data of the target object as a superposition of rigid body motion and periodic micro-motion, and using a combination of geometric primitives to represent its electromagnetic scattering center. The adjustable parameters include at least kinematic parameters, geometric parameters, and electromagnetic parameters; by changing the adjustable parameters, target object models of different types and / or different motion states can be generated.

[0009] Preferably, the target objects in S1 include one or more of the following: winged flying insects, reptiles, and stationary vertebrates; wherein, For the aforementioned flapping-wing insects, the kinematic parameters include flapping frequency and / or flapping amplitude; For the aforementioned crawling myriapods, the kinematic parameters include crawling gait frequency, stride length, and / or gait phase perturbation terms; For the aforementioned stationary vertebrates, their kinematic parameters include respiratory rate, respiratory amplitude, heart rate, and / or heart amplitude.

[0010] Preferably, the package environment model library built in S1 includes: Define the shape and size parameters of the package container; The items inside the package are defined as a randomly distributed set of point scatterers, and their statistical distribution model is stored. The statistical distribution model includes scatterer density parameters and statistical distribution parameters of radar cross section. Establish a database of complex permittivity of packaging materials in a specified frequency band.

[0011] Preferably, the radar sensor model library constructed in S1 includes: A linear frequency modulated continuous wave radar model is used as a template to define its waveform parameters, antenna parameters, and / or receiver parameters; the waveform parameters include at least the center frequency, bandwidth, and pulse repetition frequency; the antenna parameters include at least the antenna pattern model and its gain and beamwidth; the receiver parameters include at least the noise figure.

[0012] Preferably, S3 includes: At each simulation moment, the three-dimensional position and orientation of all parts of the target object are updated according to the target object motion model; The bouncing ray method is used to trace the multiple reflection, penetration and diffraction paths of discretized rays emitted from the radar in the virtual simulation scene, and the contribution of each ray's illumination area to the radar receiving antenna's scattered field is calculated based on the physical optics method. The scattered fields of all the illuminated areas are vector-superimposed to obtain the total scattered field at the corresponding simulation moment; The total scattered field is mixed with the radar transmitted signal model, and receiver noise is added to output the original radar baseband signal.

[0013] Preferably, by updating the three-dimensional position and attitude of all parts of the target object at each simulation moment and calculating the scattering field, the time-varying propagation distance caused by the periodic micro-motion of the target object is modulated into the phase change of the radar echo, thereby obtaining the micro-Doppler frequency component in the original radar baseband signal.

[0014] Preferably, S4 includes: Perform a fast Fourier transform on the original radar baseband signal in the range dimension to determine one or more range gates where the target object is located; For the signal within the selected distance gate, a time-frequency analysis is performed using a short-time Fourier transform along the time dimension to obtain a two-dimensional matrix of the signal frequency components changing over time. The amplitude of the two-dimensional matrix is ​​converted into decibel scale and then visualized to generate the time-spectrum graph.

[0015] Secondly, embodiments of the present invention provide a method for intelligent detection of a target object, comprising the following steps: S10: Acquire the radar echo signal of the target to be detected; S20: Process the radar echo signal of the target to be detected to generate the corresponding time spectrum diagram of the target to be detected; S30: Construct a dataset, including: the time-spectrum map obtained by the target object radar micro-motion signal simulation method described in the first aspect, automatically generating corresponding label information for each time-spectrum map by parsing its corresponding structured scene configuration file, and associating the time-spectrum map with the label information to form a dataset for machine learning; S40: Train a deep learning model using the dataset, input the spectrogram of the target object into the trained deep learning model, and output the target object detection result.

[0016] Preferably, S30 includes: From the structured scene configuration file, label information for machine learning tasks is extracted according to predefined mapping rules; the label information includes at least target category labels, vital sign physical parameters, and scene context information. By using a unified naming rule or index table, each of the time-spectrum images is uniquely associated with the corresponding label information to obtain sample pairs; All associated sample pairs are divided into training, validation, and test sets according to a preset ratio and stored as a dataset for machine learning.

[0017] This invention addresses the problem of detecting unusual targets in logistics security inspection scenarios. Through technological innovations such as the construction of a parameterized model library, SBR-PO hybrid physical simulation, and deep learning detection, it overcomes the bottlenecks of low computational efficiency and poor scene reproduction in traditional radar signal simulation. It provides a more efficient method for generating high-fidelity radar micro-motion signals and accurately detecting unusual targets. Specific beneficial effects are as follows: Solving the problem of data scarcity and reducing algorithm training costs: It can generate an unlimited, multi-scenario, and precisely labeled radar micro-motion signal dataset without conducting large-scale real-world tests, effectively reducing biosafety risks and significantly reducing the training cost and cycle of detection algorithms.

[0018] To improve the fidelity of simulation signals and ensure the performance of detection models: By combining the environmental characteristics of logistics parcels with the physiological rhythm characteristics of exotic pets, the SBR-PO hybrid simulation algorithm is adopted. The generated signals are highly consistent with the real security inspection scenarios. The model trained based on this data has higher detection accuracy and robustness in practical applications.

[0019] Achieve accurate identification and eliminate missed detections and misjudgments: By capturing the vital signs and movement characteristics of exotic pets, it is possible to effectively distinguish between target objects and non-target objects, and identify the target objects hidden in the package from the root.

[0020] Non-contact inspection improves security efficiency and safety: Exotic pets can be detected in packages without opening them for inspection, ensuring package integrity while improving security efficiency, reducing the risk of security personnel coming into contact with pathogens, and meeting the rapid security inspection needs of massive logistics packages. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 This is a flowchart of the target object radar micro-motion signal simulation method provided in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the generation process of the original radar baseband signal provided in this embodiment of the invention. Figure 3 This is a flowchart of the intelligent target object detection method provided in the embodiments of the present invention; Figure 4 The time-spectrum diagram of the radar micro-motion signal simulation of the flapping insect provided in the embodiment of the present invention; Figure 5 This is a diagram showing the relationship between vibration frequency, position, and time in the simulation of radar micro-motion signals of winged insects provided in this embodiment of the invention. Figure 6 This is a feature vector diagram of the radar micro-motion signal simulation of a flapping insect provided in an embodiment of the present invention; Figure 7 This is a time-spectrum diagram of the simulated micro-motion signal of vertebrate radar provided in the embodiments of the present invention; Figure 8 This is a diagram showing the relationship between vibration frequency, location, and time in a simulated vertebrate radar micro-motion signal provided in this embodiment of the invention. Figure 9 This is a feature vector diagram of the vertebrate radar micro-motion signal simulation provided in the embodiments of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] This invention discloses a method for simulating radar micro-motion signals of a target object, combined with... Figure 1 The steps are explained as follows: S1: Construct a parametric model library; the parameter space of the parametric model library includes at least a target object model library, a packaging environment model library, and a radar sensor model library; among them, the target object model library is used to store adjustable parameters of the kinematic characteristics and electromagnetic scattering characteristics of the target object; the packaging environment model library is used to store the random scattering field of the packaging environment in which the target object is located, and the electromagnetic parameters of the packaging material; the radar sensor model library is used to store the operating parameters of the radar system; S2: Based on the parametric model library, sample and combine the parameter space to generate multiple structured scene configuration files describing different virtual simulation scenarios; S3: For each scenario configuration file, a hybrid electromagnetic simulation algorithm is used to dynamically calculate the radar echo of the target object at each moment and generate the original radar baseband signal containing the micro-Doppler characteristics of the target object. S4: Perform signal processing on the original radar baseband signal to generate a time-spectrum diagram for visually displaying the micro-motion characteristics of the target.

[0025] In one embodiment, step S1 establishes a target object model library, a package and environment model library, and a radar sensor model library, respectively. For radar micro-motion detection, the kinematic characteristics and electromagnetic scattering characteristics of the target are the core factors that determine the signal characteristics. In the complex environment of the package, background clutter is determined by describing its statistical characteristics.

[0026] In logistics security inspection, target objects, package environments, and radar sensors are mathematically abstracted to establish model template libraries that can be represented by a finite number of parameters. For target objects, key physical parts that generate radar micro-motion signals are extracted, and adjustable parameters are defined for the key kinematic and electromagnetic properties of each abstract part. For package environments, the cluttered contents inside the package are treated as a random scattering field, and statistical characteristics of internal clutter are defined for different types of packages, as well as the complex permittivity for common packaging. For radar sensors, the operating parameters of the radar system are set to determine the simulation resolution and signal characteristics.

[0027] By constructing a parametric model library, motion, environmental context, and radar characteristics are all transformed into mathematically computable parametric models. The core features of all models can be flexibly adjusted and combined through parameters, providing a standardized and reusable model foundation for the generation of diverse subsequent scenarios. Furthermore, the core features of all models can be flexibly adjusted and combined through parameters, and each parameter has a clear physical meaning, providing a standardized and reusable model foundation for the generation of diverse subsequent scenarios.

[0028] In this embodiment, building the target object model library in S1 includes: The kinematic features include storing the motion data of the target object as a superposition of rigid body motion and periodic micro-motion, and using a combination of geometric primitives to represent its electromagnetic scattering center; Adjustable parameters include at least kinematic parameters, geometric parameters, and electromagnetic parameters; by changing the adjustable parameters, different types and / or different motion states of target object models can be generated.

[0029] In this embodiment, the target object mentioned in S1 includes one or more of the following: winged flying insects, myriapods, and stationary vertebrates; wherein, For flapping-wing insects, the kinematic parameters include flapping frequency and / or flapping amplitude; For crawling myriapods, the kinematic parameters include crawling cadence, stride length, and / or gait phase perturbation terms; For stationary vertebrates, their kinematic parameters include respiratory rate, respiratory amplitude, heart rate, and / or heart amplitude.

[0030] The parameters of the target object model library are shown in Table 1.

[0031] Table 1 Parameter Table of Target Object Model Library

[0032] Specifically, in constructing the target object model library, the continuous and complex motion of living organisms is decomposed into a superposition of rigid body motion and periodic micro-motions, and basic geometric primitives such as ellipsoids, cylinders, and flat plates are used to represent their main electromagnetic scattering centers. By defining the kinematic skeleton, geometric, and electromagnetic parameters, and encapsulating each parameter into a structured target description file, different types or states of insect models can be generated by changing the parameters. First, insects are initially divided into flapping-wing insects, reptilian myriapods, stationary vertebrates, and special-state categories, and corresponding motion, geometric, and electromagnetic models are constructed. For flapping-wing insects, the periodic rotation of the wings around the body joint axis causes time-varying radial velocities at various points on the wings, thus modulating the phase of the radar echo. Therefore, the complex wing motion is simplified into a simple harmonic motion model, with its angular displacement... i(t) changes with time according to formula (1).

[0033] (1) in, A Indicates the amplitude of the slapping. f Indicates the wingbeat frequency. F Indicates the initial phase.

[0034] Inside the enclosure, the limited space prevents the insects from flying freely, forcing them to engage in localized struggling movements, thus reducing their theoretical speed. v 0 by constraint factor α attenuation, α =1 indicates that it is fully restricted and introduces randomness. rand ( t ) and intermittent I ( t Simulates realistic struggling behavior. Overall translational speed. v body It is very small (close to 0), but there may be slight tremors. Therefore, the body position can be represented by formula (2). It is random jitter, and is low-frequency, small-amplitude noise.

[0035] (2) Reptilian myriapods rely on precise phase coordination of their legs for locomotion. In enclosed environments, obstructions from debris can disrupt this ideal phase relationship. f i0 Introducing the disturbance term Δ f i · or ( t To simulate this destructive effect, the leg movement phase can be represented by formula (3). The crawling speed is theoretically equal to stride length S multiplied by stride frequency. f step However, the presence of debris inside the package will reduce the actual speed, resulting in a decay function. x () quantifies this effect, and its limited crawling speed can be represented by formula (4).

[0036] (3) in, f i0 Indicates the first i The inherent phase shift of a leg, f step Δ represents step frequency, which determines the rhythm of crawling movement. f i This represents the maximum phase perturbation amplitude, simulating gait instability caused by debris inside the package. or ( t ) represents a low-frequency random perturbation function.

[0037] (4) For stationary vertebrates, respiration is not a perfect sine wave and often contains a second harmonic. The fundamental frequency plus the first harmonic is used to approximate the actual respiratory waveform to improve physiological accuracy (see formula (5)). The surface displacement waveform generated by the heartbeat is complex and contains multiple harmonics. The sum of the first three harmonics is used to approximate the heartbeat waveform, with the amplitude set at 1 / n The attenuation simulates the actual spectral characteristics, as shown in formula (6).

[0038] (5) in, A resp This indicates the maximum chest wall expansion corresponding to the main respiratory amplitude. f resp Indicates respiratory rate, A resp2 This represents the amplitude of respiratory harmonics, simulating the asymmetry between inhalation and exhalation. f This indicates the harmonic phase difference.

[0039] (6) in, A hn Indicates the first n Secondary heartbeat harmonic amplitude f heart Indicates the fundamental frequency of the heartbeat. F n This indicates the phase of each harmonic.

[0040] In the special case, the non-living target inside the package is simplified as a single-degree-of-freedom spring-mass-damped system. Environmental vibration is used as an external force input, and the system responds according to its own dynamic characteristics. Its forced vibration equation is shown in formula (7).

[0041] (7) in, m Indicates target quality. c This indicates that the damping coefficient is related to the internal friction of the material. k This indicates that the stiffness coefficient is related to the material's elasticity. A env Indicates the amplitude of environmental vibration acceleration. w env It represents the angular frequency of environmental vibration.

[0042] In general, the core parameters for flapping insects include body length (5–30 mm), wingspan (10–80 mm), wingbeat frequency (50–300 Hz), wingbeat amplitude (30–90°), skin reflectivity (0.6–0.9), and wing membrane transmissivity (0.2–0.6). For reptiles and myriapods, parameters include body length (10–50 mm), number of legs (typically 6 or 8), leg length (5–30 mm), gait frequency (1–10 Hz), stride length (2–20 mm), and skin reflectivity (0.5–0.8). For resting vertebrates, vital signs are considered, including body length (50–200 mm), body width (20–100 mm), respiratory rate (0.2–2 Hz) and amplitude (0.1–2 mm), heart rate (2–10 Hz) and amplitude (0.01–0.1 mm), and skin reflectivity (0.7–0.95). Special state classes (simulating inanimate objects) use corresponding geometric and electromagnetic parameters, and add natural frequencies (0.5–20Hz) and damping ratios (0.01–0.3) to describe their forced vibration characteristics. All parameters are given clear physical meanings and typical value ranges, and different target object models can be instantiated by changing these parameters.

[0043] In this embodiment, building the package environment model library in S1 includes: Define the shape and size parameters of the package container; The items inside the package are defined as a set of randomly distributed point scatterers, and their statistical distribution model is stored. The statistical distribution model includes the scatterer density parameter and the statistical distribution parameter of the radar cross section. Establish a database of complex permittivity of packaging materials in a specified frequency band.

[0044] Specifically, for the package and environment model library, firstly, a three-dimensional container object is defined, with its shape (e.g., cuboid, cylinder) and size as adjustable parameters, defining the boundaries of the package. All internal items and scattering bodies are constrained within this space. Secondly, the miscellaneous items inside the package, such as clothing, foam, and stuffing, are abstracted into a randomly distributed set of point scattering bodies, including their positions (…). x , y , z The scatterers are randomly distributed inside the package. The radar cross-section of each scatterer... s i Follows a log-normal distribution ln( s ) N( m , d 2 This distribution reflects the characteristic that most scatterers in a real scene are weak, while a few are strong. (Parameters) m and dIt is adjustable to simulate different types of fillers. Therefore, the scatterer can be represented by a statistical model, as shown in Equation (8).

[0045] (8) in N = r · L · W · H Indicates the number of scatterers. r This represents the scatterer density, which is the average number of scatterers per unit volume. s i Indicates the first i The radar cross section of a scatterer.

[0046] Establish a database of complex permittivity of common packaging materials such as corrugated paper, foam plastic, and bubble wrap in the millimeter-wave band. For each material, define the real part of the complex permittivity. e ’ The imaginary part of the complex permittivity affects the propagation speed of electromagnetic waves. e ’’ Reflects material loss and typical thickness. d .

[0047] All the above parameters, including container size, clutter density and distribution parameters, and packaging material type, are organized into a structured text file. Specifically, this includes the shape and size of the container (e.g., a cuboid with dimensions of 10-100 cm), and the internal random filling material is considered as a random point scattering field following a log-normal distribution (density 0.1-10 points / cm³, radar cross-section statistical parameters). m -30 to -10 dBsm d The parameters are 1-5 dB, and the complex permittivity of the packaging material (real part 1.5-3.5, imaginary part 0.01-0.5) and thickness (0.1-5 mm). This document fully defines all properties of a single package environment, making the simulation process repeatable and batch-generated.

[0048] The parameters of the package environment model library are shown in Table 2.

[0049] Table 2. Package Environment Model Library Parameter Table

[0050] In this embodiment, the construction of the radar sensor model library in S1 includes: A linear frequency modulated continuous wave radar model is used as a template to define its waveform parameters, antenna parameters, and / or receiver parameters. The waveform parameters include at least the center frequency, bandwidth, and pulse repetition frequency. The antenna parameters include at least the antenna pattern model and its gain and beamwidth. The receiver parameters include at least the noise figure.

[0051] Specifically, for the radar sensor model library, the linear frequency modulated continuous wave (FMCW) radar model, which is widely used in short-range detection, is adopted as the standard template, and the center frequency is set. f c Signal bandwidth B Frequency modulation pulse duration T c Pulse repetition frequency (PRF) and the number of pulses within a coherent processing interval (CPI). N c and frequency modulation slope K = B / T c Within a frequency-modulated pulse, the transmitted complex baseband signal is represented by a complex exponential signal with a quadratic phase change, and its instantaneous frequency increases linearly, as shown in formula (9).

[0052] (9) The spatial filtering characteristics of an antenna can simulate the change in antenna gain with the azimuth angle, approximated by a Gaussian beam model. The antenna at the azimuth angle (...) i , The power gain G( on) i , See formula (10). In SBR, the initial power weight of each emitted ray is determined by the G corresponding to its emission direction. (i , The square root of the gain determines the contribution of the returning ray, which must also be multiplied by the pattern gain of the receiving antenna during reception.

[0053] (10) in, G 0 represents the maximum gain. i 3dB and 3dB These are the half-power beamwidths for the azimuth and elevation planes, respectively.

[0054] To simulate a real radar system, the model library defines key receiver parameters, including the noise figure (NF). System noise temperature. T sys It is obtained by converting from the noise figure NF. T sys = T 0(10 NF / 10 1), of which T 0=290K is the reference temperature. Noise power. P n = k B T sys B , where k B This represents the Boltzmann constant. Complex Gaussian white noise can be generated as needed during the simulation. n ( t The real and imaginary parts are independent, and their variances are both equal. P n / 2.

[0055] All the radar system parameters mentioned above are structured and stored in a configuration file, specifically using a linear frequency modulated continuous wave radar as a template. The parameters cover waveform (center frequency 24 / 77 / 94 GHz, bandwidth 1-4 GHz, pulse repetition frequency 1-10 kHz, FM pulse duration 50 microseconds to 1 millisecond, number of pulses within the coherent processing interval 64 to 512), antenna (maximum gain 10-30 dBi, half-power beamwidth 10°-60°), receiver (noise figure 3-10 dB), and deployment geometry (detection range 0.5-5 meters). This file, combined with the environment configuration file, constitutes a complete simulation task.

[0056] The parameters of the radar sensor model library are shown in Table 3.

[0057] Table 3. Parameter Table of Package Environment Model Library

[0058] In one embodiment, S2 includes: S21: Select specific values ​​from the parameter space according to the data requirements, and write the parameters into the structured configuration file in the prescribed format. That is, the scene description file fully defines all the initial conditions and rules of this simulation.

[0059] S22: To ensure the randomness and diversity of the data, extensive sampling is performed across different parameter dimensions, and different targets, packaging, and observation conditions are combined to automatically generate a large number of different virtual scenes.

[0060] S23: It automatically generates massive and diverse simulation tasks through programs, and ensures the repeatability of the simulation process through configuration files. Through systematic parameter space exploration, it can generate data covering various simple and extreme cases, effectively simulating the uncertainty of the real world.

[0061] In practice, based on requirements, specific parameter values ​​are extracted and combined from three predefined parametric model libraries to automatically generate a large number of diverse virtual simulation scene configuration files. First, the number of scenes to be generated and the proportion of each type of target are determined, for example: 70% insects, 20% reptiles, 10% inanimate objects, and challenging scenes to be covered, such as strong clutter and thick packaging. Second, different sampling strategies are formulated: random uniform sampling, which involves completely randomizing values ​​within the allowed range of each parameter, is used to quickly generate a basic dataset and broadly explore the parameter space; targeted sampling involves dense sampling within key or extreme parameter ranges to enhance the coverage of specific challenging scenes in the dataset; and combined generation involves arranging and combining discrete parameters such as target type and packaging material type, and then combining them with the sampling results of continuous parameters to ensure scene diversity. According to the sampling strategy, a specific model is instantiated by randomly selecting a target object type according to weight, and parameters such as package size and material are determined for instantiating the packaging environment and configuring radar sensors. This process is repeated until the specified number of scenes are generated. The instantiated target object is randomly placed within the available space inside the package to ensure it does not penetrate the boundary.

[0062] Simultaneously, the package is placed in a predetermined position for rapid logical and physical plausibility checks to ensure the target is inside the package, the radar beam can illuminate the package, and signal penetration loss does not completely obliterate the echo. The rapid logical check requires the type to be less than 1 / 6 of the package's shortest side. Next, axis-aligned bounding box (AABB) detection is used to calculate the minimum bounding box of all components of the target object at the initial moment, i.e., finding its maximum and minimum values ​​in the X, Y, and Z directions. The bounding box is checked to see if it is completely within the package's internal space. The criteria are that none of the six faces of the bounding box intersect with the package's inner wall, and all vertices are inside the package; otherwise, the scene quality is marked as poor. The vector from the radar to the package's geometric center is calculated, and the angle between this vector and the radar antenna's main beam pointing vector is calculated. If this angle is greater than the antenna's half-power beamwidth... i 3dB If the package is outside the effective illumination range of the radar's main beam, the scene quality is considered poor, and it is marked. A test ray is emitted from the radar position toward the center of the package. It is only determined whether this ray intersects with any pre-set large obstruction in the scene before reaching the outer surface of the package. If it intersects, the main detection path is considered blocked, and it is marked.

[0063] Finally, discard or adjust unreasonable configurations, generate a unique configuration file for each scene that passes the verification, and organize the paths of all generated scene configuration files into one file.

[0064] In one embodiment, S3 includes: At each simulation moment, the three-dimensional position and orientation of all parts of the target object are updated according to the motion model of the target object; The bouncing ray method is used to trace the multiple reflections, penetrations and diffractions of discretized rays emitted from the radar in a virtual simulation scene, and the contribution of each ray's illumination area to the radar receiving antenna's scattered field is calculated based on physical optics. The scattered fields of all the illuminated areas are vector-superimposed to obtain the total scattered field at the corresponding simulation moment; The total scattered field is mixed with the radar transmitted signal model, and receiver noise is added to output the original radar baseband signal.

[0065] In this embodiment, by updating the three-dimensional position and attitude of all parts of the target object at each simulation moment and calculating the scattering field, the time-varying propagation distance caused by the periodic micro-motion of the target object is modulated into the phase change of the radar echo, thereby obtaining the micro-Doppler frequency component in the original radar baseband signal.

[0066] In practice, the structured scene configuration file (scene script) generated in step S2 is received. Based on classical electromagnetic theory, it numerically simulates the complete physical process of millimeter-wave radar signals from transmission and interaction with the dynamic scene to reception, outputting the original radar echo signal containing micro-motion characteristics. The bouncing ray method (SBR) is used to efficiently handle ray paths in complex environments, discretizing the electromagnetic waves emitted by the radar into millions to hundreds of millions of rays. The propagation path of each ray in the virtual scene is traced in an approximate ray form, including direct impact, reflection from object surfaces, and penetration through packaging materials. Physical optics (PO) is used to accurately calculate the scattered field of a single region. When a ray illuminates the target surface, the induced current in that surface region is calculated based on the incident electromagnetic field, and these currents are treated as new radiation sources, calculating their contribution to the radar receiving antenna. Simultaneously, based on the motion model frequency in the scene script, the precise position and attitude of all components of the target object are updated in real time. The instantaneous, dynamic geometric state is input into the SBR+PO solver to calculate the radar echo at that moment, outputting a complete complex radar signal that varies over time.

[0067] Step S2 generates the original radar baseband signal at the physical level, incorporating all physical effects such as target micro-motion, packaging penetration, multipath effects, and environmental clutter. Through temporal dynamic coupling, the movement of vital signs is directly converted into phase modulation of the radar echo signal, thereby precisely embedding micro-Doppler frequency components consistent with real physical laws into the signal. Combined with... Figure 2 As shown, the specific physical electromagnetic simulation steps are given below: First, scene analysis is performed, the scene configuration file is read, and the 3D virtual scene is accurately reconstructed, including the wrapping geometry and materials, internal random clutter scatterers, parameterized target object models, and radar sensor models. The simulation duration is then determined based on the configured time. T sim and radar pulse repetition period T r Discretize the time axis into M time points. t m ( m =0, 1, ..., M 1) At every moment t m According to the motion equations of the target object, such as the wing flapping angle i ( t m The system updates the precise three-dimensional position and orientation of all its components, including limb positions. To ensure simulation accuracy, the update frequency of the motion model should be no less than twice the radar pulse repetition frequency.

[0068] For each t m The SBR-PO hybrid algorithm is used to calculate the radar echo at that moment. Specifically, millions to hundreds of millions of discrete rays are emitted from the phase center of the radar transmitting antenna into the scene space. The power weight of each ray is determined by its angle in the antenna pattern. To balance accuracy and efficiency, an adaptive ray-tube method is employed, emitting denser rays in the main beam direction, with each ray representing a ray tube carrying energy and phase information. Acceleration is achieved using a hierarchical bounding box tree (BVH) built in real-time for dynamic scenes. Rays are first tested against the root node of the BVH, i.e., the maximum bounding box, quickly eliminating a large number of irrelevant regions and significantly reducing computational complexity. Based on geometric optics (GO) principles, the propagation path of these rays in the scene is traced one by one, simulating their reflection on the target or packaging surface, refraction and attenuation when penetrating packaging materials, and diffraction at object edges, supporting multiple reflections to simulate multipath reflection effects. During ray tracing (SBR), when a ray intersects a packaging wall, the material type is queried... e = e ’ yes ’’ Secondly, the Fresnel formula is used to calculate the reflection coefficient Γ and transmission coefficient T, and the attenuation during the transmission process is calculated. The transmitted wave amplitude is calculated according to... e αd Attenuation, where the attenuation constant α and e ’’ Relatedly, update the electric field amplitude and phase of the ray.

[0069] When rays strike the surface of a target, such as the trunk, wings, or clutter scatterer of an insect, they illuminate a small area. Let the electric field incident on this area be E. inc The unit normal vector of this surface element is n. Based on physical optics (PO) theory, the induced surface current Js is calculated from the incident electromagnetic field on this surface element, as shown in formula (11). Treating this induced current as a radiation source, the scattered field E contributing to the radar receiving antenna is calculated. scat Under the far-field approximation, for a receiving point at a distance R... ; in k=2π / λ For wave number, The unit direction vector from the surface element to the receiving point. Let be the position vector on the surface element. The scattered fields from all illuminated surfaces are vector-superimposed at the receiving antenna to obtain the value at that moment. t m Total scattered field E total ( t m ). Characterizing spatial phase delay, electromagnetic waves require phase accumulation to propagate from a surface element to a receiving point. kR This is a fundamental characteristic of wave phenomena, indicating that the phase of a wave lags linearly with the propagation distance.

[0070] Characterizing the direction of radiation, the path difference from different surface elements to the receiving point is different. When summing over all surface elements in the integral, this factor determines the interference pattern of the scattered field.

[0071] When an incident electromagnetic wave strikes the surface of a target, it induces a current on the surface. The current density at the induced surface is determined by the unit normal vector and the magnetic field strength of the incident magnetic field. At each point, it is proportional to the magnetic field strength, while the unit normal vector is determined by the shape. It is twice the magnetic field strength because an ideal conductor surface will completely reflect the electromagnetic waves, equivalent to the superposition of the incident and reflected waves in the millimeter-wave radar band, with wavelengths of only 3-4 mm. Most biological tissues exhibit high reflectivity and low transmission of electromagnetic waves, making them suitable for the application range of the above formula.

[0072] (11) in, For the incident magnetic field, or For free space wave impedance, The unit direction vector of the incident wave.

[0073] Because the target is at different times t m At different locations, the echo relative to the radar propagation distance R( tm The distance change caused by the micro-motion is time-varying. This distance change is directly converted into the phase change of the echo signal, as shown in formula (12). l For wavelength. By each t m Update the geometry and calculate the scattering field, taking the periodic motion R( of vital signs) as an example. t m The phase directly and precisely modulated onto the radar echo ( t m This allows for the implantation of real micro-Doppler frequency components into the signal. .

[0074] (12) The calculated time-series complex scattering field E total ( t m The signal is then conjugate-mixed (i.e., descrambled) with a radar transmit signal model such as FMCW, and then low-pass filtered to output a complex baseband signal sequence. s bb [ n The signal already contains all the deterministic phase modulations generated by the interaction between the target's micro-motions and the environment. Subsequently, based on the noise model defined in the radar sensor model library, the corresponding complex Gaussian white noise can be generated. n [ n And superimposed on s bb [ n This allows us to obtain a signal that more closely resembles the actual output of a receiver. s [ n ]= s bb [ n ]+ n [ n ].

[0075] In one embodiment, step S4 processes and transforms the original radar baseband signal generated in S3, which is not directly interpretable by the human eye (i.e., the original time-domain voltage signal), into a two-dimensional image—a time-spectrum graph—that can clearly and intuitively display the micro-motion characteristics of the target. This transforms the micro-Doppler information implicit in the one-dimensional time series into an image format suitable for processing by deep learning models such as convolutional neural networks (CNNs). First, a Fast Fourier Transform (FFT) is performed on the echo within each radar pulse, separating echoes with different time delays (corresponding to different distances) into different range gates. Second, for range gates that may contain targets, a Short-Time Fourier Transform (STFT) is used to analyze the signal along the radar pulse sequence. Through a sliding time window, the frequency components of the signal within that short time period are continuously calculated. The amplitude of the STFT result is converted to a decibel scale and presented as an image, i.e., a time-spectrum graph.

[0076] This step effectively extracts and condenses the information most relevant to target detection by focusing on the range gate that may contain the target and performing time-frequency analysis. It explicitly transforms the micro-Doppler features implicit in the one-dimensional time-domain signal into visual patterns in the two-dimensional image.

[0077] In this embodiment, S4 includes: Perform a fast Fourier transform on the original radar baseband signal in the range dimension to determine one or more range gates where the target object is located; For the signal within the selected distance gate, a time-frequency analysis is performed using short-time Fourier transform along the time dimension to obtain a two-dimensional matrix of the signal frequency components changing with time. The amplitude of the two-dimensional matrix is ​​converted into decibel scale and then visualized to generate a time-frequency spectrum.

[0078] The specific steps for radar signal processing and time-spectrum graph generation are given below: By utilizing the wide bandwidth of radar signals, echoes at different distances are separated in terms of energy, thereby isolating scattered signals from different depth regions within the envelope into different range gates. First, the echo signals received within each radar transmit pulse or FMCW frequency-modulated pulse cycle are... s chirp ( t Perform a Fast Fourier Transform (FFT). Based on the time-shifting characteristics of the Fourier Transform, echoes with different time delays (i.e., different distances) will exhibit spectral lines with different peak values ​​in the frequency domain. The amplitude spectrum of the transformed result is shown in the figure. | This constitutes a one-dimensional range profile, where each frequency unit (bin) corresponds to a specific range interval, called a range gate. Based on the known package center distance R in the dynamic scene configuration file... pkg and its depth in the radar line-of-sight direction D pkgDirectly calculate the interval occupied by the package in the distance dimension. R pkg D pkg / 2, R pkg + D pkg / 2], and map this interval to the corresponding range cell on the radar range image, thereby determining the range gate range where the target is located and selecting the range gate.

[0079] Within a selected range gate, the instantaneous changes in the signal frequency components are analyzed along the time axis to extract the time-varying Doppler frequency generated by the target's micro-motion. f d ( t This refers to the micro-Doppler characteristic. A short-time Fourier transform (STFT) is used, through a sliding time window. w ( t The signal is segmented and its spectrum is calculated to obtain a two-dimensional representation of the signal's frequency components over time. For discrete signals, a signal sequence with a selected distance gate is used. x [ m ] (length is M), a window with a step size R and a length of L is used to perform FFT on each window segment, see formula (13).

[0080] (13) in, i For time frame indexing, k This is a frequency index. Results X [ k , i ] is a two-dimensional complex matrix.

[0081] Secondly, the amplitude or power of the STFT result is calculated and converted to a decibel (dB) scale to enhance visual contrast, as shown in formula (14). P [ k , i The matrix is ​​treated as a color image. The image is standardized and then appropriately cropped, scaled, or interpolated to generate a time-spectrum image of fixed size. Figure 5 and Figure 8 In the diagram, the horizontal axis represents time, the vertical axis represents Doppler frequency, and the brightness of a pixel represents the intensity of that frequency component at that moment.

[0082] (14) in, e It is a very small value to prevent taking the logarithm of zero.

[0083] Based on the same inventive concept, embodiments of the present invention also provide a method for intelligent detection of target objects, combined with Figure 3 The steps are explained as follows: S10: Acquire the radar echo signal of the target to be detected; S20: Process the radar echo signal of the target to be detected and generate the corresponding time spectrum diagram of the target; S30: Construct a dataset, including: time-spectrum maps obtained by the radar micro-motion signal simulation method for target objects based on the same concept; automatically generate corresponding label information for each time-spectrum map by parsing its corresponding structured scene configuration file; and associate the time-spectrum map with the label information to form a dataset for machine learning. S40: Train a deep learning model using the dataset, input the spectrogram of the target object into the trained deep learning model, and output the target object detection result.

[0084] In one embodiment, S30 automatically associates the time-spectrum graph generated in the preceding steps with its corresponding precisely known scene physical parameters to construct a large-scale, high-quality standard dataset that can be directly used for machine learning training. This includes: From the structured scene configuration file, according to predefined mapping rules, extract label information for machine learning tasks; the label information includes at least the target category label, vital sign physical parameters, and scene context information; By using a unified naming rule or index table, each time spectrum image is uniquely associated with its corresponding label information to obtain sample pairs; All associated sample pairs are divided into training, validation, and test sets according to a preset ratio and stored as a dataset for machine learning.

[0085] In this embodiment, the label information also includes target information, motion and physical characteristics, signal quality and difficulty, and machine learning auxiliary format. A time-spectrum graph, representing each specific data point, is associated and paired with its corresponding automatically generated label file. This ultimately generates a usable standard dataset that can be directly imported into mainstream machine learning frameworks such as TensorFlow and PyTorch for training, validating, and testing various detection algorithms. This solves the fundamental problems of high cost, low efficiency, and large subjective errors in real-world data annotation.

[0086] In step S30, features are extracted from each time-spectrum image to construct sample pairs. The specific steps are as follows: First, the time-frequency spectrum is summed in both the full frequency and full time dimensions to obtain the time and frequency profiles, as follows: Figure 4~6The example shown illustrates the specific process of wing flapping, a vital sign, by simulating the generation of a certain wing-flapping insect. The structured scene configuration parameters before the simulation are: SNR 19.0dB, distance 1.6 m, and angle -21.6°. These parameters are derived from the structured scene configuration file before the simulation and are directly defined by parameters such as radar geometry and receiver noise. When building the dataset, these parameters are automatically extracted by parsing the configuration file and associated with the images as labels. Figure 4 The indices in the time-frequency spectrum diagram indicate the frequency channel numbers after the Short-Time Fourier Transform (STFT), with each channel representing a physical frequency width Δ. f The sampling rate and FFT points during simulation determine the system frequency; in sample 2, the system is set to 2Hz. Find the index of the energy peak in the frequency profile column; then, the wingbeat frequency = index × Δ f (2 × 87.5 = 175 Hz). The micro-motion amplitude is obtained by multiplying the peak-valley fluctuation amplitude of the time profile by the system calibration parameters. This calibration coefficient is derived from the radar equations and signal processing link in the simulation scenario configuration file. Figure 4 The red curve reflects the continuity of the target's movement, and smooth fluctuations indicate a stable movement state. The peak of the blue curve in the 40–60 range corresponds to the gait frequency, which is a typical movement frequency characteristic of flapping-winged insects. Figure 5 The diagram shows the relationship between the vibration frequency, location, and time of a flapping-winged insect, illustrating the intermittent flapping motion.

[0087] like Figure 7~9 The example shown is a simulation of a vertebrate species. The structured scene configuration parameters before the simulation are: SNR 12.6dB, distance 2.7 m, and angle -28.3°. These parameters are derived from the structured scene configuration file before the simulation and are directly defined by parameters such as radar geometry and receiver noise. When building the dataset, these parameters are automatically extracted by parsing the configuration file and associated with the images as labels. Figure 7 The time-spectrum plot shown is used to visually display the variation pattern of micro-Doppler frequency over time. The time profile is used to extract the normalized peak-to-valley difference of the micro-motion amplitude and further convert it into actual displacement through calibration coefficients. The frequency profile is used to locate the energy peak index and multiply it by the frequency resolution Δ output from the configuration file. f Obtain the precise frequency. Figure 7 The horizontal axis index represents the spectrum sampling point number, which needs to be linearly converted to the actual frequency using the radar system's sampling frequency and the number of FFT points. The frequency value in the sample label is the final converted result: frequency = index × Δ. f。

[0088] Figure 7The red curve reflects the continuity of the target's movement; smooth fluctuations indicate a stable movement state. Vertebrates involve both respiratory and cardiac movements, unlike insects which exhibit a single high-frequency wingbeat, hence the two peaks in their respiratory movements. Respiration is a slow, large-amplitude, periodic movement with a relatively low frequency in vertebrates. Figure 7 In the diagram, the peak of the blue curve in the 0–20 Hz range indicates that the energy of the respiratory signal is concentrated in this frequency band, reflecting the slow, periodic respiratory movements of vertebrates. This peak corresponds to the weak energy in the high-frequency band of the radar spectrum, matching the high-frequency micro-motion characteristics of the heartbeat, resulting in a calculated respiratory frequency of 1.7 Hz for the sample label. The heartbeat is a periodic pulsation that is faster and has a smaller amplitude than respiration. After radar phase modulation, physical simulation, and time-frequency processing, its micro-motion energy is concentrated in the 40–60 Hz range, forming a narrower, slightly weaker peak, reflecting the rapid, periodic cardiac pulsation of vertebrates. This peak represents the strongest low-frequency component in the spectrum, matching the dominant micro-motion characteristics of respiration, resulting in a calculated respiratory frequency of 5.0 Hz for the sample label. Figure 8 A diagram showing the location and time relationship between respiratory and heart rate in vertebrates, illustrating respiratory and heartbeat movements.

[0089] Secondly Figure 6 and Figure 9 These are all feature vector maps, which are numerical features further reduced in dimensionality from the time-frequency spectrogram, and serve as input to the deep learning model. Dimensionality reduction of the time-frequency spectrogram extracts the most discriminative core features, forming a one-dimensional feature vector that is then visualized. The curve represents the dimensionality reduction result of the high-dimensional time-frequency features; the peak in the 150–200 range is a characteristic feature of myriapods and can be used for target classification and recognition.

[0090] Finally, based on the configuration file, corresponding category labels and simulation parameter annotations are added to each set of feature vectors, and integrated into a structured numerical dataset, realizing the transformation of time spectrum plots into a computable and modelable dataset.

[0091] In step S40, each scene configuration file is a description containing all the physical states of the scene. From this configuration file, semantic information and physical features meaningful to the machine learning task are extracted according to predefined rules. Example of label content: Classification label target_type:winged_insect (target object category); physical attributes vital_sign_frequency: [150.0, 0.8] (wing flapping frequency Hz, breathing frequency Hz), target_avgrcs: -25.3 (target's average radar cross section, dBsm), snr_estimated: 18.5 (estimated signal-to-noise ratio, dB); scene context package_material: cardboard, angle_of_arrival: 45.0 (degrees). All extracted label information is organized into a standard format compatible with the machine learning framework. Each time-spectrum image file is uniquely and definitively associated with its corresponding label file using a unified naming rule or index table. Secondly, the paired data is quickly validated to ensure that the spectrograms and labels are consistent in quantity and dimension. All spectrogram-label pairs are divided into three sets according to a predetermined ratio: 70% training set, 15% validation set, and 15% test set. The sets are then packaged into a standard dataset format, such as a format that can be directly read by TensorFlow's tf.data.Dataset or PyTorch's Dataset class.

[0092] Finally, the entire parameter space sampling records and all scene configuration files are archived to construct a queryable metadata database. Researchers can flexibly reconstruct or subset the dataset according to physical conditions.

[0093] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0094] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for simulating radar micro-motion signals of a target object, characterized in that, Includes the following steps: S1: Construct a parametric model library; the parameter space of the parametric model library includes at least a target object model library, a packaging environment model library, and a radar sensor model library; wherein, the target object model library is used to store adjustable parameters of the kinematic characteristics and electromagnetic scattering characteristics of the target object; the packaging environment model library is used to store the random scattering field of the packaging environment in which the target object is located, and the electromagnetic parameters of the packaging material; the radar sensor model library is used to store the operating parameters of the radar system; S2: Based on the parameterized model library, sample and combine the parameter space to generate multiple structured scene configuration files describing different virtual simulation scenarios; S3: For each scenario configuration file, a hybrid electromagnetic simulation algorithm is used to dynamically calculate the radar echo of the target object at each moment, and generate the original radar baseband signal containing the micro-Doppler characteristics of the target object. S4: Perform signal processing on the original radar baseband signal to generate a time-spectrum diagram for visually displaying the micro-motion characteristics of the target.

2. The target object radar micro-motion signal simulation method as described in claim 1, characterized in that, The construction of the target object model library in S1 includes: The kinematic features include storing the motion data of the target object as a superposition of rigid body motion and periodic micro-motion, and using a combination of geometric primitives to represent its electromagnetic scattering center. The adjustable parameters include at least kinematic parameters, geometric parameters, and electromagnetic parameters; by changing the adjustable parameters, target object models of different types and / or different motion states can be generated.

3. The target object radar micro-motion signal simulation method as described in claim 1, characterized in that, The target objects described in S1 include one or more of the following: flapping-winged insects, myriapods, and stationary vertebrates; wherein: For the aforementioned flapping-wing insects, the kinematic parameters include flapping frequency and / or flapping amplitude; For the aforementioned crawling myriapods, the kinematic parameters include crawling gait frequency, stride length, and / or gait phase perturbation terms; For the aforementioned stationary vertebrates, their kinematic parameters include respiratory rate, respiratory amplitude, heart rate, and / or heart amplitude.

4. The target object radar micro-motion signal simulation method as described in claim 1, characterized in that, The package environment model library built in S1 includes: Define the shape and size parameters of the package container; The items inside the package are defined as a randomly distributed set of point scatterers, and their statistical distribution model is stored. The statistical distribution model includes scatterer density parameters and statistical distribution parameters of radar cross section. Establish a database of complex permittivity of packaging materials in a specified frequency band.

5. The target object radar micro-motion signal simulation method as described in claim 1, characterized in that, The radar sensor model library built in S1 includes: A linear frequency modulated continuous wave radar model is used as a template to define its waveform parameters, antenna parameters, and / or receiver parameters; the waveform parameters include at least the center frequency, bandwidth, and pulse repetition frequency; the antenna parameters include at least the antenna pattern model and its gain and beamwidth; the receiver parameters include at least the noise figure.

6. The target object radar micro-motion signal simulation method as described in claim 1, characterized in that, S3 includes: At each simulation moment, the three-dimensional position and orientation of all parts of the target object are updated according to the target object model library; The bouncing ray method is used to trace the multiple reflection, penetration and diffraction paths of discretized rays emitted from the radar in the virtual simulation scene, and the contribution of each ray's illumination area to the radar receiving antenna's scattered field is calculated based on the physical optics method. The scattered fields of all the illuminated areas are vector-superimposed to obtain the total scattered field at the corresponding simulation moment; The total scattered field is mixed with the radar transmitted signal model, and receiver noise is added to output the original radar baseband signal.

7. The target object radar micro-motion signal simulation method as described in claim 6, characterized in that, By updating the three-dimensional position and attitude of all parts of the target object at each simulation moment and calculating the scattered field, the time-varying propagation distance caused by the periodic micro-motion of the target object is modulated into the phase change of the radar echo, and the micro-Doppler frequency component in the original radar baseband signal is obtained.

8. The target object radar micro-motion signal simulation method as described in claim 1, characterized in that, S4 include: Perform a fast Fourier transform on the original radar baseband signal in the range dimension to determine one or more range gates where the target object is located; For the signal within the selected distance gate, a time-frequency analysis is performed using a short-time Fourier transform along the time dimension to obtain a two-dimensional matrix of the signal frequency components changing over time. The amplitude of the two-dimensional matrix is ​​converted into decibel scale and then visualized to generate the time-spectrum graph.

9. A method for intelligent detection of a target object, characterized in that, Includes the following steps: S10: Acquire the radar echo signal of the target to be detected; S20: Process the radar echo signal of the target to be detected to generate the corresponding time spectrum diagram of the target to be detected; S30: Constructing a dataset, including: obtaining the time-spectrum map based on the target object radar micro-motion signal simulation method according to any one of claims 1 to 8, automatically generating corresponding label information for each time-spectrum map by parsing its corresponding structured scene configuration file, and associating the time-spectrum map with the label information to form a dataset for machine learning; S40: Train a deep learning model using the dataset, input the spectrogram of the target object into the trained deep learning model, and output the target object detection result.

10. The intelligent target object detection method as described in claim 9, characterized in that, S30 includes: From the structured scene configuration file, label information for machine learning tasks is extracted according to predefined mapping rules; the label information includes at least target category labels, vital sign physical parameters, and scene context information. By using a unified naming rule or index table, each of the time-spectrum images is uniquely associated with the corresponding label information to obtain sample pairs; All associated sample pairs are divided into training, validation, and test sets according to a preset ratio and stored as a dataset for machine learning.

Images