Low-altitude target identification system and method combined with multi-band signal processing

Abstract

The invention relates to the technical field of target detection and identification, and discloses a low-altitude target identification system and method combined with multi-band signal processing. Comprising the steps of establishing a cross-frequency implicit vector model and constraining a geometric scale, a scattering center and rotor harmonic waves through synchronous acquisition of an L-band radar, a Ku-band radar and a millimeter-wave radar, and realizing unified characterization of multi-frequency signals; receiving a global satellite navigation system reflection signal, communication channel state information or a very high frequency / ultrahigh frequency passive radar signal to form prior discrimination, and triggering high-frequency radar short-time imaging; carrying out propagation correction on millimeter wave, terahertz and infrared signals in combination with a meteorological profile and generating an environment label; determining a sampling window by using the region of interest and rotor phase prediction, and optimizing a residence strategy; harmonic waves are extracted from radar and acoustic signals, robust pairing is carried out, and a cross-domain consistency score output category is calculated. The method realizes multi-frequency multi-source cooperative processing, has environment self-adaption and closed-loop feedback capabilities, and can improve the precision and stability of low-altitude target identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target detection and identification, and particularly relates to a low-altitude target identification system and method combining multi-band signal processing. BACKGROUND

[0002] Low-altitude airspace activities include small unmanned aerial vehicles, birds, and ultra-light aircrafts. Such targets are small in size, low in flight altitude, and strong in maneuverability, and are often mixed with buildings, ground clutter, or natural bird flocks in the atmospheric boundary layer and complex terrain background. The clutter and target signals are often difficult to separate in the low-altitude scene, which may cause false alarms and missed alarms.

[0003] In the prior art, common solutions include multi-frequency radar detection, passive radar monitoring, and auxiliary identification using satellite navigation reflection signals or communication signals. Multi-frequency radars have complementary advantages in detection distance and resolution, but the physical characteristics of signals at different frequencies differ significantly. Existing methods mostly stop at result fusion and lack physical consistency modeling, which has limited stability in complex environments. Passive monitoring methods have good concealment, but can only provide limited prior information and are difficult to support reliable identification alone. On the other hand, weather conditions have a significant impact on the propagation of millimeter waves, terahertz, and infrared waves, and existing methods generally lack effective propagation correction mechanisms, and the performance decreases significantly in fog, precipitation, and high humidity environments. In addition, in multi-target scenarios, existing scheduling strategies lack consideration of sampling time and resource allocation, which may lead to track loss or identification delay. For rotorcraft targets, although harmonic characteristics exist in radar and acoustic signals, existing methods mostly analyze in a single signal domain, and the cross-domain matching and mutual verification mechanism is insufficient. Once a signal channel degrades or some harmonics are missing, the identification accuracy will be significantly affected. Therefore, we propose a low-altitude target identification system and method combining multi-band signal processing. SUMMARY

[0004] To overcome the deficiencies of the prior art, the present application provides a low-altitude target identification method combining multi-band signal processing to solve the technical problems existing in the prior art.

[0005] The above technical purpose of the present application is achieved by the following technical scheme:

[0006] A low-altitude target identification method combining multi-band signal processing, comprising the following steps:

[0007] A) multi-frequency hidden vector modeling step: at least two of L-band radar, Ku-band radar and millimeter wave radar are used to synchronously collect low-altitude targets, a cross-frequency hidden vector model is established, and the geometric scale consistency, scattering center sparsity and rotor harmonic spacing consistency of the targets are constrained to realize unified representation of signals of different frequency bands;

[0008] B) passive prior trigger step: at least two of global satellite navigation system reflected signals, fifth generation mobile communication or wireless fidelity channel state information, very high frequency or ultra-high frequency passive radar signals are received, coherence time, channel matrix dispersion and frequency difference characteristics are extracted, and prior discrimination of target existence and type is formed; when the prior discrimination reaches a preset threshold, Ku-band radar or millimeter wave radar is triggered to perform short-time imaging collection;

[0009] C) atmospheric profile correction step: based on real-time meteorological profile, temperature, humidity and refractive index structure parameters are obtained, and propagation correction and deconvolution processing are performed on the echo signals of millimeter wave radar, terahertz channel and infrared channel to generate multi-modal signals fused by cross-frequency consistency constraint, and an environment label is output;

[0010] D) residence scheduling step: L-band radar is used to maintain long-range track tracking, based on the interest area result of step B) and the environment label of step C), combined with target rotor phase prediction, the best recognition time window is determined, Ku-band radar is controlled to perform short-time sampling in the window, and optimization is performed among detection probability, track continuity and micro-Doppler signal-to-noise ratio to obtain a dynamic residence strategy;

[0011] E) cross-domain harmonic pairing step: rotor harmonic families are extracted from the corrected radar time-frequency spectrum and array microphone sound spectrum, when there is noise interference or some harmonics are missing, neighbor number interpolation or energy proportion constraint is used to realize robust pairing, harmonic number consistency relationship is established and cross-domain consistency score is calculated, and the target category is output;

[0012] The interest area result of step B) is input into step D);

[0013] The correction result and environment label of step C) are input into step A) and step E);

[0014] The identification result and consistency score of step E) are fed back to step B) and step D) to form a closed-loop adaptive identification process.

[0015] Preferably, the multi-frequency radar sampling clock is synchronized based on a common time reference, and the synchronization accuracy is better than one fifth of the sampling period of each radar system;

[0016] The cross-frequency latent vector model is trained through few-shot metric learning to quickly establish feature metric standards for new target categories under low data conditions, supporting the rapid expansion and recognition of new target categories.

[0017] The latent vector output in step A) is configured as the prior weight for the cross-domain harmonic pairing consistency calculation in step E), which is used to enhance the pairing confidence of the same source harmonics and suppress the heterogeneous interference term.

[0018] Preferably, the reflected signals from the global satellite navigation system are received using a combination of right-hand circular polarization and linear polarization to enhance the coherent detection capability for low-altitude small targets;

[0019] The wireless fidelity channel state information constitutes a micro-motion discrimination index through the eigenvalue dispersion and rank variation of the subcarrier correlation matrix.

[0020] The very high frequency or ultra-high frequency passive radar acquires time difference and frequency difference information through multi-baseline deployment to form a coarse trajectory;

[0021] Furthermore, step B) employs a layered triggering logic, first performing single-channel discrimination on each of the aforementioned passive channels, and then performing multi-channel joint discrimination based on the single-channel discrimination results. When the joint discrimination conditions are met, the millimeter-wave radar short-time imaging acquisition is triggered.

[0022] Preferably, the meteorological profile source includes any one or more of microwave radiometer data, radiosonde data, or numerical weather analysis.

[0023] When visibility is low or humidity is too high based on the meteorological profile, the weight of the terahertz or infrared channel is reduced, and only the millimeter wave channel is retained to participate in multimodal signal fusion.

[0024] Preferably, the target rotor phase prediction is obtained by estimating the target's historical micro-Doppler period to accurately determine the optimal identification time window.

[0025] When the risk of track loss exceeds a preset threshold, the short-time sampling duty cycle is reduced to prioritize track continuity.

[0026] Preferably, in step E), when the confidence of the output target category is low, on the one hand, step B) is requested to increase the weight of the passive signal by lowering the trigger threshold or increasing the sampling duration, and on the other hand, step C) is requested to increase the spatiotemporal resolution of atmospheric correction to verify the discrimination.

[0027] The cross-domain consistency score output in step E) is used to update the residency policy priority queue in step D), and the output target category label is used as feedback input to step B) to optimize the prior triggering logic for the next cycle.

[0028] A low-altitude target identification system combining multi-band signal processing includes:

[0029] a multi-frequency radar subsystem, corresponding to method step A), for collecting L-band radar, Ku-band radar and millimeter-wave radar signals, establishing a cross-frequency steering vector model, and performing cross-frequency consistency constraints based on geometric scaling, scattering center sparsity and rotor harmonic spacing;

[0030] a passive perception subsystem, corresponding to method step B), for receiving global navigation satellite system reflection signals, fifth-generation mobile communication or wireless fidelity channel state information, very high frequency or ultra-high frequency passive radar signals, and performing prior trigger discrimination;

[0031] a meteorological acquisition subsystem, corresponding to method step C), for acquiring meteorological profiles and generating propagation correction parameters and environmental labels;

[0032] a scheduling control subsystem, corresponding to method step D), for maintaining track following, determining an optimal recognition time window according to a region of interest, an environmental label and a rotor phase prediction, controlling short-time sampling and optimizing a residence strategy;

[0033] an acoustic perception subsystem, corresponding to method step E), for acquiring array microphone signals, extracting rotor harmonic families and performing robust pairing in a missing harmonic scenario, generating cross-domain consistency scores and class recognition results;

[0034] a processor and a memory for storing and executing program instructions, the processor updating a trigger condition of the passive perception subsystem and a residence strategy of the scheduling control subsystem based on the cross-domain consistency scores.

[0035] Preferably, the multi-frequency radar subsystem comprises,

[0036] a time synchronization module for aligning radar samples of each system to a common time reference;

[0037] the passive perception subsystem comprises:

[0038] a global navigation satellite system reflection receiving module using right-hand circular polarization and linear polarization joint reception;

[0039] a communication receiving module supporting subcarrier level channel state information sampling;

[0040] a multi-baseline very high frequency or ultra-high frequency passive radar receiving module.

[0041] Preferably, the meteorological acquisition subsystem comprises a microwave radiometer module and a numerical weather analysis interface module;

[0042] the scheduling control subsystem comprises a high-speed trigger interface connected with the multi-frequency radar subsystem, for performing short-time imaging collection within an optimal time window of rotor phase.

[0043] A priority queue module is used to output the cross-domain consistency score update sampling priority of step E) of the method.

[0044] Preferably, the acoustic perception subsystem is a uniform linear array or a planar array, the inter-element spacing of which is set according to the target main radiation frequency band, and the category confidence output by the acoustic perception subsystem is input into the processor as a feedback signal.

[0045] The processor is further configured to adjust the trigger threshold of the passive perception subsystem based on the category confidence.

[0046] When the category confidence is insufficient, the meteorological acquisition subsystem is requested to increase the spatiotemporal resolution of atmospheric correction.

[0047] In summary, the present application has the following advantages:

[0048] By collecting multi-frequency radar data under a unified time reference and introducing geometric scaling, scattering distribution and rotor harmonic constraints in the latent vector space, the observation results of different frequency bands can be mapped to the same feature space. This approach avoids classification bias caused by frequency band differences, making the recognition result more reliable and facilitating the rapid introduction of new target categories in the case of limited sample size; By using global satellite navigation system reflection signals, wireless local area network or fifth generation mobile communication channel state information, and very high frequency / ultra high frequency passive radar signals, the existence of the target can be determined first, and then the short-time acquisition of the library wave band or millimeter wave radar is triggered. This passive judgment and active follow-up method reduces the invalid residence time of active radar, reduces energy consumption and electromagnetic exposure, and concentrates limited sampling resources in more valuable spatiotemporal regions;

[0049] In view of the influence of fog, haze, precipitation and humidity on millimeter wave, terahertz and infrared channels, the system corrects the propagation process through meteorological profile data and generates an environmental label to guide the weight adjustment of different channels. For example, when the humidity is high, the proportion of terahertz and infrared channels is reduced, and only millimeter wave is retained as the main channel, thereby ensuring the stability of the recognition performance in bad weather; By predicting the rotor phase to determine the short-time sampling window, the micro-Doppler features collected are more clear. In the case of multiple targets, a priority queue is introduced to arrange the residence order according to the confidence of passive discrimination, the environmental label and the cross-domain harmonic consistency score. In this way, the recognition accuracy and track continuity can be considered, and the target loss risk can be reduced; When extracting rotor harmonics from radar signals and acoustic signals, there may be noise or partial loss. Through group sparsity modeling and neighbor number interpolation methods, stable cross-domain harmonic pairing can be achieved, and the credibility of the pairing can be improved combined with the latent vector weight. The final consistency score not only serves as the basis for the recognition result, but also is fed back to the passive trigger and scheduling link to adjust the subsequent sampling strategy, thereby forming an adaptive closed-loop mechanism. Attached Figure Description

[0050] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0051] 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.

[0052] Example 1

[0053] refer to Figure 1 A low-altitude target identification system and method combining multi-band signal processing, comprising the following steps:

[0054] A) Multi-frequency latent vector modeling steps: Simultaneously acquire low-altitude targets using at least two of the following systems: L-band radar, Ku-band radar, and millimeter-wave radar. Establish a cross-frequency latent vector model and constrain the consistency of the target's geometric scale scaling, the sparsity of the scattering center, and the consistency of the rotor harmonic spacing to achieve a unified representation of signals in different frequency bands.

[0055] B) Passive prior triggering steps: Receive at least two of the following: global satellite navigation system reflected signals, fifth-generation mobile communication or wireless fidelity channel state information, and very high frequency or ultra-high frequency passive radar signals; extract coherence time, channel matrix dispersion and frequency difference features to form prior discrimination of target existence and type; when the prior discrimination reaches a preset threshold, trigger Ku-band radar or millimeter-wave radar to perform short-time imaging acquisition.

[0056] C) Atmospheric stratification correction steps: Based on real-time meteorological profiles, temperature, humidity and refractive index structural parameters are obtained. The echo signals from millimeter-wave radar, terahertz channel and infrared channel are propagation corrected and deconvolved to generate multi-mode signals fused by cross-frequency consistency constraints, and environmental labels are output.

[0057] D) Dwelling and scheduling steps: Use L-band radar to maintain long-range track tracking. Based on the region of interest results in step B) and the environmental labels in step C), combined with target rotor phase prediction, determine the optimal identification time window. Within this window, control Ku-band radar to perform short-term sampling and optimize the detection probability, track continuity and micro-Doppler signal-to-noise ratio to obtain a dynamic dwelling strategy.

[0058] E) Cross-domain harmonic pairing step: Extract the rotor harmonic family in the corrected radar time-frequency spectrum and array microphone sound spectrum, when there is noise interference or partial harmonic missing, use adjacent number interpolation or energy proportion constraint to realize robust pairing, establish the harmonic number consistency relationship and calculate the cross-domain consistency score, output the target category;

[0059] Wherein, the interest area result of step B) is input into step D);

[0060] The correction result of step C) and the environment label are input into step A) and step E);

[0061] The identification result and consistency score of step E) are fed back to step B) and step D) to form a closed-loop adaptive identification process.

[0062] System composition and deployment

[0063] Multi-frequency radar subsystem: L-band, Ku-band, millimeter wave radar, all of which can operate independently, and the sampling end has a unified time reference.

[0064] Passive perception subsystem: global navigation satellite system (GNSS) reflection reception, wireless fidelity (Wi-Fi) / fifth generation mobile communication (5G) channel state information (CSI) reception, very high frequency (VHF) / ultra high frequency (UHF) passive radar (PBR) multi-baseline reception.

[0065] Weather acquisition subsystem: microwave radiometer / sonde / numerical weather product interface; connect terahertz detector and infrared detector in parallel to provide terahertz / infrared channel signals.

[0066] Scheduling control subsystem: responsible for short-time sampling triggering within the phase window, high-speed interface, priority queue management.

[0067] Acoustic perception subsystem: array microphone (linear array or surface array), set the element spacing according to the target main radiation frequency band.

[0068] Processor and memory: execute A-E step procedures of the embodiment to complete closed-loop update.

[0069] The sampling clock of the three-mode radar is aligned to a unified time reference (such as PTP), and the cross-mode time synchronization error δt≤T s / 5, T s is the self-sampling period of each mode;

[0070] Synchronized echo discrete sequence: y f [n], f∈{L, Ku, mm}.

[0071] Unified latent vector and physical regular inclusion

[0072] Let the shared latent vector Physical generation sub-operator H f (·) Establish observation consistency:

[0073]

[0074] In the formula, H f Integrating antenna pattern, bandwidth / pulse parameters, target equivalent scatterer and micro-Doppler modulation;

[0075] Three regularities are consistent dimensions:

[0076]

[0077]

[0078] Unified optimization:

[0079]

[0080] In the formula S f Extracting hidden components related to geometric profiles, α is the cross-frequency geometric scale ratio; W is the scattering center sparse transform, Δf k is the kth harmonic spacing, is the mean value; Alternating least squares or Adam: generate H f (z) Then calculate the gradient, and then update z until convergence.

[0081] Few-shot metric learning:

[0082]

[0083] c k : class prototype; d(·): Euclidean / Mahalanobis / cosine distance; α ∈ (0, 1).

[0084] After training, new categories only need a small amount of sample updates c k can be identified.

[0085] Exported as prior weight for the E-step harmonic number k pair weighting.

[0086] 1) More stable boundary recognition: Even if individual frequency bands are disturbed, the hidden space is still jointly constrained by multiple frequencies and will not fluctuate with a single path;

[0087] 2) Low cost of new class expansion: A small amount of samples can build new class prototypes, adapting to the rapid evolution of models in the field;

[0088] 3) More reliable cross-domain pairing: ​As E-step weights, can suppress acoustic / radar single-domain abnormal peaks, and improve the success rate of one-to-one correspondence of harmonics.

[0089] GNSS reflection:

[0090]

[0091] In the formula, τ c is the coherence time, is the mirror / diffuse ratio, κ polarization consistency (right circular polarization + linear polarization joint), and σ is the Sigmoid.

[0092] Wi-Fi / 5G CSI:

[0093] p CSI = σ (b1Δλ + b2r (t) + b3ACF (f0))

[0094] In the formula, Δλ = (λ max - λ min ) / ∑λ k , r(t) is the time-varying rank, and ACF(f0) is the base frequency autocorrelation peak;

[0095] VHF / UHF PBR:

[0096] p PBR = σ (c1SNR + c2fit (TDOA, FDOA))

[0097] Hierarchical triggering and region of interest (ROI):

[0098] P joint = σ (w1p GNSS + w2p CSI + w3p PBR -T joint )

[0099] When P joint ≥ 0.5:

[0100] Trigger Ku / millimeter wave to conduct short-time imaging sampling;

[0101] Generate ROI: (azimuth, elevation, range gate, radial velocity gate).

[0102] Through the passive first and then active passive link with wide coverage, good concealment and low cost to filter the space-time worth watching;

[0103] With the following effects:

[0104] Resource concentration: active residence only occurs in high-value areas, improving the proportion of effective samples under the same sampling budget;

[0105] Concealed security: reduce the duty cycle and peak-to-average power ratio when actively transmitting, suitable for sensitive area deployment;

[0106] Robust to complex environments: multi-channel comprehensive discrimination can correct each other, reducing the false trigger of single pointer.

[0107] Propagation kernel and deconvolution

[0108] T, ρ from the meteorological profile v , Propagation kernel of frequency band f:

[0109]

[0110] Deconvolution and cross-frequency consistency fusion:

[0111]

[0112] Solve FISTA / ADMM.

[0113] Environmental labels and adaptive weights

[0114] Output Env = {VIS, RH, Rain, ; C_n^2} and other environmental labels.

[0115] When VIS is low or RH is high.

[0116] w THz ,w IR ↓,w mm ↑

[0117] Only keep the dominant fusion of millimeter wave weights; push the label to the D step synchronously;

[0118] Translate weather parameters into propagation availability, first physically correct distortion, then use labels to guide subsequent strategies;

[0119] Fog / haze / rain stability: terahertz / infrared is easily affected by humidity and scattering, adaptive weights can avoid dragging behind the channel to lower the overall quality;

[0120] Cross-frequency fusion is reliable: first do physical correction and then fusion, avoid treating propagation distortion as a feature, reduce misjudgment;

[0121] Parameters are interpretable: environmental labels can be used for threshold care and quality control tracing on the operation side, making it easy to quickly define problems on site.

[0122] Optimal time window driven by phase

[0123] Estimate the fundamental frequency f0(t) and angular velocity ω(t) = 2πf0(t) from historical micro-Doppler, phase

[0124] Select window center:

[0125]

[0126] where, is the stripe clarity (harmonic separability, SNR, bandwidth ratio, etc.), sampling window [t ＊ -Δ, t ＊ +Δ].

[0127] Cost function and priority queue;

[0128] Online cost:

[0129]

[0130] Multi-target priority:

[0131] Priority = a·P joint (B) + b·EnvScore(C) + c·CrossScore(E)

[0132] Assign Ku-band time window and duty cycle to different targets by queue; if the risk of losing the track exceeds the threshold, reduce the duty cycle and prioritize tracking;

[0133] Quantify when to look, how long to look, and who to look at first into an optimization problem, and use phase predictability and prior scores to schedule;

[0134] Multi-target throughput improvement: priority queue allows resources to flow to "more likely to give a clear conclusion" targets first, and the overall recognition throughput is higher;

[0135] Trajectory continuity guarantee: when tracking is unstable, automatically yield the active position, reducing the situation of "recognition is good for a while, and the track is all up to collision";

[0136] End-to-end closed loop: queue score accesses the output of B / C / E, and scheduling is based on the dynamic changes of the comprehensive state of what is seen, what is calculated, and what is transmitted well.

[0137] Group sparse pairing and missing robustness;

[0138] Peak set

[0139] Group sparse target:

[0140]

[0141] where, is the cross-domain same-order coefficient, Ψ miss Use adjacent order interpolation / energy ratio to handle noise or partial harmonic missing situations;

[0142] Consistency score and classification;

[0143]

[0144] If CrossScore ≥ θ c , output class and confidence; if acoustic SNR is too low, downgrade to cross-frequency radar harmonic consistency discrimination (only use , still use

[0145] Low confidence callback;

[0146] If confidence is lower than τ p :

[0147] Lower trigger threshold or increase passive sampling duration;

[0148] Increase atmospheric correction spatio-temporal resolution;

[0149] Update queue priority and window with the latest CrossScore;

[0150] Mutually verify radar and acoustic with physically-motivated harmonic order one-to-one correspondence; suppress false peaks when mismatched;

[0151] Anti-noise and anti-missing: even if wind noise / masking causes partial disappearance of acoustic harmonics, group sparsity and neighboring order interpolation can still recover the matching relationship;

[0152] End correctable: low confidence is not a direct rejection, but triggers re-sampling, re-correction, and re-scheduling to improve the convergence probability of difficult samples;

[0153] Degradation without disability: when the acoustic link degrades, the system automatically switches to cross-frequency radar consistency to maintain basic recognition, so as not to collapse the entire link.

[0154] Closed-loop linkage and engineering parameters;

[0155] Closed-loop path: steps B to D (ROI / confidence); steps C to A / E (clean domain and environmental label); steps A to E (prior weight); steps E to B / D (callback / queue); steps D to A / E (high-quality short sampling feedback).

[0156] Hidden vector dimension D = 64-256; λ 1,2,3 ∈ [10 -3 , 10 -1 ];

[0157] Trigger threshold T joint Optimized with development set; phase window half-width Δ = 0.1-0.2 rotor periods;

[0158] Group sparse coefficient [η, ξ ∈ [10-4 ,10 -2 ];σ f Take target base frequency 5% - 10%.

[0159] Any signal quality / discriminant confidence of any step can affect the next round of acquisition and correction through a queue and threshold loop, and the system can self-adjust with environmental and target changes.

[0160] Deployable and maintainable: each indicator and threshold can be exposed, facilitating on-site monitoring, A / B testing, and problem tracing.

[0161] Realistic constraints: explicitly incorporate engineering constraints such as weather, phase, energy consumption / duty cycle, and concurrent targets into formulas and strategies, and balance recognition rate and stability under realistic constraints.

[0162] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for low-altitude target identification combining multi-band signal processing, characterized in that, Includes the following steps: A) Multi-frequency latent vector modeling steps: Simultaneously acquire low-altitude targets using at least two of the following systems: L-band radar, Ku-band radar, and millimeter-wave radar. Establish a cross-frequency latent vector model and constrain the consistency of the target's geometric scale scaling, the sparsity of the scattering center, and the consistency of the rotor harmonic spacing to achieve a unified representation of signals in different frequency bands. B) Passive prior triggering steps: Receive at least two of the following: global satellite navigation system reflected signals, fifth-generation mobile communication or wireless fidelity channel state information, and very high frequency or ultra-high frequency passive radar signals; extract coherence time, channel matrix dispersion and frequency difference features to form a priori judgment of target presence and type; when the prior judgment reaches a preset threshold, trigger Ku-band radar or millimeter-wave radar to perform short-time imaging acquisition. C) Atmospheric stratification correction steps: Based on real-time meteorological profiles, temperature, humidity and refractive index structural parameters are obtained. The echo signals from millimeter-wave radar, terahertz channel and infrared channel are propagation corrected and deconvolved to generate multi-mode signals fused by cross-frequency consistency constraints, and environmental labels are output. D) Dwelling and scheduling steps: Use L-band radar to maintain long-range track tracking. Based on the region of interest results in step B) and the environmental labels in step C), combined with target rotor phase prediction, determine the optimal identification time window. Within this window, control Ku-band radar to perform short-term sampling and optimize between detection probability, track continuity and micro-Doppler signal-to-noise ratio to obtain a dynamic dwelling strategy. E) Cross-domain harmonic pairing steps: Extract rotor harmonic families from the calibrated radar time spectrum and array microphone sound spectrum. When there is noise interference or some harmonics are missing, robust pairing is achieved by using adjacent number interpolation or energy proportion constraint. Establish harmonic number consistency relationship and calculate cross-domain consistency score, and output target category. The region of interest results from step B) are input into step D). The correction results of step C) and the environmental label input in steps A) and E). The identification results and consistency scores of step E) are fed back to steps B) and D) to form a closed-loop adaptive identification process.

2. The low-altitude target identification method combining multi-band signal processing according to claim 1, characterized in that, The multi-frequency radar sampling clock is synchronized based on a common time reference, and the synchronization accuracy is better than one-fifth of the sampling period of each radar system. The cross-frequency latent vector model is trained through few-shot metric learning to quickly establish feature metric standards for new target categories under low data conditions, supporting the rapid expansion and recognition of new target categories. The latent vector output in step A) is configured as the prior weight for the cross-domain harmonic pairing consistency calculation in step E), which is used to enhance the pairing confidence of the same source harmonics and suppress the heterogeneous interference term.

3. The low-altitude target identification method combining multi-band signal processing according to claim 2, characterized in that, The Global Navigation Satellite System uses a combination of right-hand circular polarization and linear polarization to receive reflected signals in order to enhance the coherent detection capability of small targets at low altitudes. The wireless fidelity channel state information constitutes a micro-motion discrimination index through the eigenvalue dispersion and rank variation of the subcarrier correlation matrix. The very high frequency or ultra-high frequency passive radar acquires time difference and frequency difference information through multi-baseline deployment to form a coarse trajectory; Furthermore, step B) employs a layered triggering logic, first performing single-channel discrimination on each of the aforementioned passive channels, and then performing multi-channel joint discrimination based on the single-channel discrimination results. When the joint discrimination conditions are met, the millimeter-wave radar short-time imaging acquisition is triggered.

4. The low-altitude target identification method combining multi-band signal processing according to claim 3, characterized in that, The meteorological profile sources include any one or more of microwave radiometers, radiosonde data, or numerical weather analysis. When visibility is low or humidity is too high based on the meteorological profile, the weight of the terahertz or infrared channel is reduced, and only the millimeter wave channel is retained to participate in multimodal signal fusion.

5. The low-altitude target identification method combining multi-band signal processing according to claim 4, characterized in that, The target rotor phase prediction is obtained by estimating the target's historical micro-Doppler period to accurately determine the optimal identification time window. When the risk of track loss exceeds a preset threshold, the short-time sampling duty cycle is reduced to prioritize track continuity.

6. The low-altitude target identification method combining multi-band signal processing according to claim 5, characterized in that, In step E), when the confidence of the output target category is low, on the one hand, step B) is requested to increase the weight of the passive signal by reducing the trigger threshold or increasing the sampling duration, and on the other hand, step C) is requested to increase the spatiotemporal resolution of atmospheric correction to verify the discrimination. The cross-domain consistency score output in step E) is used to update the residency policy priority queue in step D), and the output target category label is used as feedback input to step B) to optimize the prior triggering logic for the next cycle.

7. A low-altitude target identification system combining multi-band signal processing, characterized in that, include: The multi-frequency radar subsystem, corresponding to step A of the method, is used to collect signals from L-band radar, Ku-band radar and millimeter-wave radar, establish a cross-frequency latent vector model, and perform cross-frequency consistency constraints based on geometric scale scaling, scattering center sparsity and rotor harmonic spacing. The passive sensing subsystem (corresponding to step B) is used to receive reflected signals from the global satellite navigation system, channel status information of fifth-generation mobile communication or wireless fidelity, and very high frequency or ultra-high frequency passive radar signals, and to perform prior trigger discrimination. The meteorological acquisition subsystem (corresponding to method step C) is used to acquire meteorological profiles and generate propagation correction parameters and environmental labels. The scheduling and control subsystem (corresponding to step D) is used to maintain track tracking and determine the optimal identification time window based on the region of interest, environmental labels, and rotor phase prediction, control short-term sampling, and optimize the dwell strategy. The acoustic perception subsystem (corresponding to method step E) is used to acquire array microphone signals, extract rotor harmonic families and perform robust pairing in scenarios with missing harmonics, and generate cross-domain consistency scores and category recognition results. A processor and memory are provided for storing and executing program instructions. The processor updates the triggering conditions of the passive sensing subsystem and the residency policy of the scheduling control subsystem based on the cross-domain consistency score.

8. A low-altitude target identification system combining multi-band signal processing according to claim 7, characterized in that, The multi-frequency radar subsystem includes, The time synchronization module is used to align the samples from radars of different systems to a common time reference; The passive sensing subsystem includes: A global satellite navigation system reflection receiver module employing a combination of right-hand circular polarization and linear polarization reception; A communication receiving module that supports subcarrier-level channel state information sampling; Multi-baseline VHF or UHF passive radar receiver module.

9. A low-altitude target identification system combining multi-band signal processing according to claim 7, characterized in that, The meteorological acquisition subsystem includes a microwave radiometer module and a numerical weather analysis interface module; The scheduling and control subsystem includes a high-speed trigger interface connected to the multi-frequency radar subsystem, used to perform short-time imaging acquisition within the optimal time window of the rotor phase; The priority queue module is used to update the sampling priority based on the cross-domain consistency score output in step E) of the method.

10. A low-altitude target identification system combining multi-band signal processing according to claim 7, characterized in that, The acoustic sensing subsystem is a uniform linear array or area array, with the array element spacing set according to the target main radiation frequency band, and its output category confidence is used as a feedback signal input to the processor. The processor is further configured to: adjust the trigger threshold of the passive sensing subsystem based on the category confidence level; When the confidence level of the category is insufficient, the meteorological acquisition subsystem is requested to increase the spatiotemporal resolution of atmospheric correction.

Images