An unmanned aerial vehicle signal detection and storage method and system based on multi-dimensional feature joint discrimination

Abstract

The application discloses a kind of based on multi-dimension feature joint discrimination unmanned aerial vehicle signal detection warehousing method and system, specifically is: wideband electric detection receiver gathers original electromagnetic spectrum signal and constructs original spectrum dataset;Five-layer quantization algorithm screening system is constructed, in turn through adaptive noise floor estimation, signal-to-noise ratio quantization decision, spectral feature euclidean distance matching, known signal feature correlation coefficient discrimination, multi-dimensional hash deduplication, environmental noise floor, clutter interference, known signal and repeated signal are gradually excluded, and retain strange unmanned aerial vehicle effective radiation signal;Extract time-frequency feature, modulation feature, intra-pulse feature and other parameters and encapsulate metadata;Spectrum feature fingerprint is written into feature library and indexed on site.The present application realizes the accurate sorting, adaptive discrimination and self-iterative library building of unmanned aerial vehicle signal in complex electromagnetic environment by multi-level mathematical model and multi-feature fusion decision, and significantly improves the continuous perception ability of anti-unmanned aerial vehicle system.

Description

Technical Field

[0001] This invention relates to the fields of electronic reconnaissance spectrum sensing, electromagnetic spectrum signal sorting of low-altitude UAVs, joint identification of multi-dimensional features in complex electromagnetic environments, and self-iterative storage of anti-UAV spectrum feature databases. Specifically, it relates to a method for detecting and storing UAV signals based on joint identification of multi-dimensional features. Background Technology

[0002] Currently, with the continuous surge in the number of low-altitude consumer drones, modified non-standard drones, and illegal drones flying without authorization, the electromagnetic environment in urban areas, border ports, and key locations is becoming increasingly complex. Background noise, power frequency harmonics, ground communication, broadcast radar, and sudden clutter are intertwined and superimposed, making it increasingly difficult to sort the electromagnetic spectrum signals of drones.

[0003] Existing anti-drone electronic reconnaissance and spectrum sensing systems rely heavily on pre-stored known drone spectrum feature templates for matching and identification. They can only effectively sort drone models that have been filed and finalized in the database. For unfamiliar drone radiation signals that appear for the first time, operate under entirely new conditions, or are modified non-standard, the feature database has no prior templates, and traditional equipment lacks an effective feature identification and sorting mechanism, making it impossible to automatically separate effective target signals.

[0004] Current engineering applications face significant technical bottlenecks:

[0005] (1) Traditional spectrum acquisition adopts a full-disk blind acquisition and storage mode, without quantization and modelable hierarchical identification logic. The raw data is mixed with a large amount of environmental noise and various interferences, and cannot automatically purify the effective signals of unfamiliar UAVs.

[0006] (2) Existing signal sorting mostly uses fixed empirical thresholds, which do not have the ability to adapt to the environment and iteratively change. They cannot adapt to changes in the electromagnetic environment of different regions, which can easily lead to the false rejection of weak target signals or the failure to filter out clutter interference.

[0007] (3) Lacking a multi-dimensional feature joint identification architecture, relying solely on a single frequency point or energy feature for judgment, the sorting robustness is poor, and it is difficult to distinguish complex UAV signals with the same frequency but different systems or different frequencies but the same features.

[0008] (4) Without a screening-packaging-on-site storage closed-loop mechanism, unfamiliar drone signals are only stored offline as raw logs. Feature modeling and template archiving are not completed. When the same type of drone approaches again, the system still cannot identify and sort it, thus losing continuous perception capability.

[0009] To address the shortcomings of existing technologies, there is an urgent need to design a method for detecting and storing drone signals in a database, in order to solve the bottleneck problems of difficulty in identifying unfamiliar drones, weak on-site documentation capabilities, and lack of adaptive iteration capabilities in existing technologies. Summary of the Invention

[0010] The purpose of this invention is to provide a method for detecting and storing UAV signals based on multi-dimensional feature joint identification. Through multi-level mathematical models and multi-feature fusion judgment, it can filter out background noise, interference, and known signals step by step, accurately select unfamiliar UAV signals, and build a database on site, meeting the application needs of low-altitude security and key area air defense projects.

[0011] The technical solution to achieve the purpose of this invention is: a method for detecting and storing drone signals based on multi-dimensional feature joint discrimination, comprising the following steps:

[0012] Step S1, Acquisition of raw electromagnetic spectrum signals: Broadband spectrum interception is performed on the monitored airspace through a broadband electronic reconnaissance receiver, and raw electromagnetic spectrum signals are acquired in real time. The output includes time-domain discrete sequence, frequency-domain FFT sequence, IQ sampling data and basic signal characteristic parameters, and the raw spectrum dataset is constructed.

[0013] Step S2, Multi-dimensional feature joint identification and five-layer quantization screening: Construct a multi-dimensional feature joint identification and five-layer quantization algorithm screening system, which sequentially uses adaptive noise floor estimation algorithm, signal-to-noise ratio quantization decision algorithm, spectrum feature Euclidean distance matching algorithm, known signal library feature correlation coefficient identification algorithm, and multi-dimensional hash deduplication algorithm to eliminate environmental noise floor, clutter interference, known communication radar signals, archived UAV signals and duplicate collected signals, retaining only the effective radiation signals of unknown UAVs that have not been archived;

[0014] Step S3, Multi-dimensional feature extraction and metadata encapsulation: For unfamiliar UAV signals that have passed the joint screening of multi-dimensional features, extract multi-dimensional parameters such as time-frequency features, modulation features, and intra-pulse features, and complete the data format normalization and integrated encapsulation of the acquisition time, space, orientation, and scene metadata.

[0015] Step S4, on-site real-time storage and subsequent signal sorting and identification: The packaged unfamiliar UAV spectrum feature fingerprint is written into the local spectrum feature library on-site, and a new exclusive template index entry is added; when the unfamiliar UAV of the same type reappears in the reconnaissance airspace, multi-dimensional feature matching and sorting are completed based on the template in the library to achieve accurate target identification and situation warning.

[0016] Furthermore, in step S1, a discrete-time signal sequence is obtained by continuously sampling the monitored spatial domain through a broadband radio frequency receiving front-end. , The frequency domain spectrum sequence is obtained by FFT transformation. ;

[0017] The fundamental characteristic parameters of a signal include its center frequency. Bandwidth usage The parameters include signal arrival time (TOA), signal pulse width (PW), signal amplitude, time-domain energy, and spectral envelope shape.

[0018] Furthermore, in step S2, the adaptive noise floor estimation algorithm of the first layer uses a sliding window recursive mean model to estimate the noise power, calculated according to the following formula:

[0019]

[0020] in, For the first Point real-time noise power estimate, For smooth forgetting factor, The amplitude of the signal in the time domain;

[0021] Set the noise floor threshold ,in This is the noise margin coefficient, when If the noise is determined to be environmental background noise, it will be directly removed; otherwise, it will be retained and proceed to the next level of multi-dimensional feature identification.

[0022] Further, in step S2, the signal-to-noise ratio quantization decision algorithm of the second layer calculates the signal-to-noise ratio according to the following formula. :

[0023]

[0024] in The effective frequency band of the signal; The frequency band is the pure noise range; N and M are the number of signal and noise sampling points, respectively. For the first The spectral amplitude value of each frequency point;

[0025] Set low threshold High threshold ;:

[0026] when Invalid weak signals are directly removed.

[0027] hour The time stamp is questionable and needs further verification;

[0028] when Then proceed to the next layer of multi-dimensional feature joint identification.

[0029] Furthermore, in step S2, the Euclidean distance matching algorithm for the spectral features of the third layer is executed as follows:

[0030] Constructing a 5-dimensional normalized spectral feature vector ,in For the normalized position of the spectral peak, For 3dB normalized bandwidth, For the spectral roll-off factor, For sidelobe attenuation, It is the time-frequency energy entropy;

[0031] Calculate the Euclidean distance between the eigenvector of the signal under test and the standard eigenvector of known clutter in the library:

[0032]

[0033] In the formula, The feature vector of the signal to be measured. The standard eigenvectors of known clutter within the library;

[0034] Set distance threshold ,like If it is determined to be clutter, it will be removed; otherwise, it will be retained as a suspicious drone target signal and allowed to proceed to the next layer.

[0035] Furthermore, in step S2, the known signal library feature correlation coefficient discrimination algorithm of the fourth layer adopts the Pearson correlation coefficient. Calculate the similarity between the current signal and known signals in the database using the following formula:

[0036]

[0037] In the formula, This is the current signal feature vector. The feature vectors of known signals within the library;

[0038] Set the relevant threshold as ,when Signals that are known to have been archived are directly removed; only those that are retained are removed. The system identifies unfamiliar and unknown drone signals and performs system-level feature identification.

[0039] Furthermore, in step S2, the multidimensional hash deduplication algorithm of the fifth layer is executed as follows:

[0040] Combined center frequency Bandwidth usage Collection timestamp eigenvectors Jointly generate hash value The generated hash value is compared with the local hash table. If they match, the signal is considered a duplicate and is removed. If they do not match, the signal is confirmed as a new, unfamiliar, and valid drone signal.

[0041] Furthermore, in step S3, the extracted time-frequency features include signal time-frequency distribution map features and instantaneous frequency change features; the extracted modulation features include modulation scheme, symbol rate, frequency hopping period, and frequency hopping sequence; the extracted intra-pulse features include intra-pulse coding scheme and intra-pulse modulation parameters.

[0042] Data format normalization includes uniform sampling rate, uniform data bit width, and uniform IQ storage format;

[0043] Metadata includes the collection time (UTC time), GPS latitude and longitude coordinates, azimuth angle, and airspace scene label.

[0044] A UAV signal detection and database entry system based on multi-dimensional feature joint discrimination is provided to implement the aforementioned UAV signal detection and database entry method based on multi-dimensional feature joint discrimination, comprising:

[0045] The signal acquisition module is used to intercept the broadband spectrum of the monitored airspace through the broadband radio frequency receiving front end, acquire the original electromagnetic spectrum signal on site in real time, and output the time-domain discrete sequence, frequency-domain FFT sequence, IQ sampling data and basic signal characteristic parameters.

[0046] The multi-level joint screening module is used to perform multi-dimensional feature joint screening and five-level quantization algorithm screening. This module includes an adaptive noise floor estimation sub-module, a signal-to-noise ratio quantization decision sub-module, a spectrum feature Euclidean distance matching sub-module, a known signal library feature correlation coefficient screening sub-module, and a multi-dimensional hash deduplication sub-module connected in sequence.

[0047] The feature extraction and encapsulation module is used to extract time-frequency features, modulation features, intra-pulse features and multi-dimensional parameters from the filtered unfamiliar drone signals, perform data format normalization and bind the acquired spatiotemporal, azimuth and scene metadata;

[0048] The spectrum feature library entry module is used to write the packaged unfamiliar UAV spectrum feature fingerprint into the local spectrum feature library and create index entries.

[0049] The signal matching and recognition module is used to perform multi-dimensional feature matching and sorting based on templates in the library when reproducing similar UAVs in the future, so as to achieve target recognition and situational warning.

[0050] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the UAV signal detection and database entry method based on multi-dimensional feature joint discrimination.

[0051] Compared with the prior art, the significant advantages of this invention are:

[0052] (1) This invention constructs a five-layer joint discrimination algorithm system that includes adaptive noise floor estimation, signal-to-noise ratio energy decision, spectral feature Euclidean distance metric, known signal correlation coefficient discrimination and multidimensional hash deduplication. This system achieves the step-by-step filtering of environmental noise floor, clutter interference and known signals. It can accurately remove the effective signals of unknown UAVs from complex electromagnetic environments. Compared with the existing decision methods that rely on a single frequency point or energy feature, the sorting robustness is significantly improved.

[0053] (2) The present invention adopts an adaptive sliding window recursive mean noise estimation algorithm, which can dynamically adjust the decision threshold according to the real-time electromagnetic environment, has adaptive environment iteration capability, can adapt to the electromagnetic environment changes in different regions, and avoids the problem of weak target signal being mistakenly eliminated or clutter interference being missed due to fixed empirical thresholds.

[0054] (3) This invention constructs multidimensional normalized spectral feature vectors and matches them with Euclidean distance, and combines Pearson correlation coefficient to identify the system dimension. It can effectively distinguish complex UAV signals with the same frequency but different system and different frequency but the same feature, which significantly improves the accuracy of signal sorting.

[0055] (4) The present invention establishes a closed-loop mechanism of screening-packaging-on-site storage. After multi-dimensional feature extraction and normalization packaging, the spectrum fingerprint of unfamiliar UAV signals is written into the feature library on-site and an index is established, completing feature modeling and template filing. When the same type of UAV approaches again, the system can quickly identify and sort it, realizing continuous perception and self-iterative library building capabilities. Attached Figure Description

[0056] Figure 1 This is a flowchart of the UAV signal detection and database entry method based on multi-dimensional feature joint identification according to the present invention. Detailed Implementation

[0057] This invention provides a method for detecting and storing drone signals based on multi-dimensional feature joint discrimination, including:

[0058] Step S1, Acquisition of raw electromagnetic spectrum signals: Broadband spectrum interception is performed on the monitored airspace through a broadband electronic reconnaissance receiver, and raw electromagnetic spectrum signals are acquired in real time. The output includes time-domain discrete sequence, frequency-domain FFT sequence, IQ sampling data and basic signal characteristic parameters, and the raw spectrum dataset is constructed.

[0059] Step S2, Multi-dimensional feature joint identification and five-layer quantization screening: Construct a multi-dimensional feature joint identification and five-layer quantization algorithm screening system, which sequentially uses adaptive noise floor estimation algorithm, signal-to-noise ratio quantization decision algorithm, spectrum feature Euclidean distance matching algorithm, known signal library feature correlation coefficient identification algorithm, and multi-dimensional hash deduplication algorithm to eliminate environmental noise floor, clutter interference, known communication radar signals, archived UAV signals and duplicate collected signals, retaining only the effective radiation signals of unknown UAVs that have not been archived;

[0060] Step S3, Multi-dimensional feature extraction and metadata encapsulation: For unfamiliar UAV signals that have passed the joint screening of multi-dimensional features, extract multi-dimensional parameters such as time-frequency features, modulation features, and intra-pulse features, and complete the data format normalization and integrated encapsulation of the acquisition time, space, orientation, and scene metadata.

[0061] Step S4, on-site real-time storage and subsequent signal sorting and identification: The packaged unfamiliar UAV spectrum feature fingerprint is written into the local spectrum feature library on-site, and a new exclusive template index entry is added; when the unfamiliar UAV of the same type reappears in the reconnaissance airspace, multi-dimensional feature matching and sorting are completed based on the template in the library to achieve accurate target identification and situation warning.

[0062] Furthermore, in step S2, the multi-dimensional feature joint screening five-layer quantization algorithm specifically includes:

[0063] The first layer: adaptive sliding window background noise screening, using recursive mean to estimate noise power in real time, and eliminating pure environmental noise based on adaptive decision threshold.

[0064] Adaptive background noise estimation uses a sliding window recursive mean model:

[0065]

[0066] In the formula: For the first Point real-time noise power estimate, For smooth forgetting factor, The amplitude of the signal in the time domain;

[0067] Set the noise floor threshold , This is the noise margin coefficient, when If identified as environmental background noise, it is directly removed; otherwise, it is retained and proceeds to the next level of multi-dimensional feature identification.

[0068] The second layer: Signal-to-noise ratio (SNR) dual-threshold quantization screening, which calculates the SNR by integrating the frequency band energy, and sets high and low thresholds to eliminate invalid weak signals and mark suspicious signals.

[0069] The formula for calculating the signal-to-noise ratio quantization is:

[0070]

[0071] in: The effective frequency band of the signal; The frequency band is the pure noise range; N and M are the number of signal and noise sampling points, respectively.

[0072] Set low threshold High threshold ; Directly eliminate invalid weak signals. The mark is questionable and awaits review. Proceed to the next layer of multi-dimensional feature joint identification.

[0073] The third layer: Euclidean distance matching and screening of spectral features, constructing a multi-dimensional normalized spectral feature vector, calculating the spatial distance with the standard clutter feature library, and eliminating non-target clutter signals with similar shapes.

[0074] Constructing a 5-dimensional normalized spectral feature vector Includes spectral peak position, 3dB bandwidth, spectral roll-off coefficient, sidelobe attenuation, and time-frequency entropy; Euclidean distance calculation formula:

[0075]

[0076] In the formula, The feature vector of the signal to be measured. The standard eigenvectors of known clutter within the library;

[0077] set up If it is a distance threshold, if Signals identified as clutter or non-target signals are discarded; those with a distance greater than a threshold are retained as suspicious UAV target signals and proceed to the subsequent multi-dimensional feature identification process.

[0078] The fourth layer involves filtering based on the known signal database. This involves calculating the correlation coefficient between the signal to be tested and known communication, radar, and archived UAV signals in the database. Signals with high correlation are directly removed, while low-correlation, unfamiliar, and unknown signals are retained.

[0079] Known signal feature correlation coefficient discrimination formula:

[0080]

[0081] In the formula, The feature vector of the signal to be measured. The feature vectors of known UAV signals within the database;

[0082] Set the relevant threshold as When the correlation coefficient Signals identified as known and already documented are directly removed; only those already retained are retained. The system identifies unfamiliar and unknown drone signals and performs system-level feature identification.

[0083] The fifth layer: multi-dimensional hash deduplication screening, which integrates frequency point, bandwidth, timestamp, and feature vector to generate a unique hash value, eliminates duplicate signals with the same target, and completes multi-dimensional feature joint identification closed-loop screening.

[0084] Multidimensional hash deduplication uses center frequency point ,bandwidth timestamp eigenvectors Jointly generate hash value The hash table is compared with the local hash table, and if a match is found, it is considered a duplicate signal and is removed.

[0085] This invention constructs a multi-dimensional feature joint discrimination algorithm system, integrating adaptive background noise estimation, signal-to-noise ratio energy judgment, spectral feature measurement, known signal correlation coefficient discrimination, and multi-level hash deduplication quantitative screening logic to filter out environmental interference step by step and accurately identify unknown UAV signals. Through multi-dimensional feature extraction and parameter normalization in the time-frequency domain and modulation domain, it enables real-time on-site database entry and archiving of the spectral fingerprint of unfamiliar UAVs. During subsequent signal reproduction, multi-dimensional matching is performed based on templates in the database, achieving rapid sorting, identification, and situational awareness. The algorithm is adaptively iterative, adaptable to complex electromagnetic environments, and has strong engineering feasibility.

[0086] Furthermore, in step S1, a discrete-time signal sequence is obtained by continuously sampling the monitored spatial domain through a broadband radio frequency receiving front-end. , The frequency domain spectrum sequence is obtained by FFT transformation. Synchronously collect IQ data and center frequency. Bandwidth usage Basic parameters such as temporal energy and spectral envelope constitute the original input dataset for multi-dimensional feature discrimination.

[0087] Furthermore, in step S1, the fundamental characteristic parameters of the signal include the signal center frequency. Bandwidth usage The parameters include signal arrival time (TOA), signal pulse width (PW), signal amplitude, time-domain energy, and spectral envelope shape.

[0088] Furthermore, in step S3, for the signal that has passed the multi-dimensional feature joint screening, the extracted time-frequency features include signal time-frequency distribution map features and instantaneous frequency change features; the modulation features include modulation scheme, symbol rate, frequency hopping period, and frequency hopping sequence; and the intra-pulse features include intra-pulse coding scheme and intra-pulse modulation parameters.

[0089] Furthermore, in step S3, data format normalization includes uniform sampling rate, uniform data bit width, and uniform IQ storage format; metadata includes acquisition time, GPS latitude and longitude coordinates, azimuth angle, and airspace scene label.

[0090] Furthermore, in step S4, the spectrum feature library establishes a joint index of frequency points and multi-dimensional features; when the same type of UAV re-enters the country, the Euclidean distance and correlation coefficient are reused to complete multi-dimensional feature matching, thereby realizing automatic sorting, identity recognition, target trajectory association and low-altitude situational alarm.

[0091] This invention also provides a UAV signal detection and database entry system based on multi-dimensional feature joint discrimination, used to implement the aforementioned UAV signal detection and database entry method based on multi-dimensional feature joint discrimination, including:

[0092] The signal acquisition module is used to intercept the broadband spectrum of the monitored airspace through the broadband radio frequency receiving front end, acquire the original electromagnetic spectrum signal on site in real time, and output the time-domain discrete sequence, frequency-domain FFT sequence, IQ sampling data and basic signal characteristic parameters.

[0093] The multi-level joint screening module is used to perform multi-dimensional feature joint screening and five-level quantization algorithm screening. This module includes an adaptive noise floor estimation sub-module, a signal-to-noise ratio quantization decision sub-module, a spectrum feature Euclidean distance matching sub-module, a known signal library feature correlation coefficient screening sub-module, and a multi-dimensional hash deduplication sub-module connected in sequence.

[0094] The feature extraction and encapsulation module is used to extract time-frequency features, modulation features, intra-pulse features and multi-dimensional parameters from the filtered unfamiliar drone signals, perform data format normalization and bind the acquired spatiotemporal, azimuth and scene metadata;

[0095] The spectrum feature library entry module is used to write the packaged unfamiliar UAV spectrum feature fingerprint into the local spectrum feature library and create index entries.

[0096] The signal matching and recognition module is used to perform multi-dimensional feature matching and sorting based on templates in the library when reproducing similar UAVs in the future, so as to achieve target recognition and situational warning.

[0097] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the UAV signal detection and database entry method based on multi-dimensional feature joint discrimination.

[0098] The present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments.

[0099] Example

[0100] Combination Figure 1 This embodiment provides a method for detecting and storing drone signals based on multi-dimensional feature joint identification, specifically including the following steps:

[0101] Step 1: Acquisition of raw electromagnetic spectrum signals;

[0102] By continuously sampling the monitored spatial domain through a broadband radio frequency receiving front-end, a discrete-time signal sequence is obtained: , The frequency domain spectrum sequence is obtained by FFT transformation: Synchronously collect IQ data and center frequency. Bandwidth usage Basic parameters such as temporal energy and spectral envelope constitute the original input dataset for multi-dimensional feature discrimination.

[0103] Step 2: Multi-dimensional feature joint screening with five-layer quantification for progressive selection;

[0104] First layer: Adaptive sliding window background noise filtering

[0105] The local noise power is estimated in real time using a recursive forgetting mean algorithm.

[0106]

[0107] In the formula, the smooth forgetting factor Noise margin coefficient Noise floor threshold Judgment Rules: Determined as ambient background noise, it was directly eliminated; The candidates are retained and proceed to the next level of multi-dimensional screening.

[0108] Second layer: Signal-to-noise ratio (SNR) quantization dual threshold screening

[0109] Signal-to-noise ratio is calculated using bandwidth energy integration:

[0110]

[0111] Project Pre-planning , , Weak and invalid signals should be removed. The label is questionable; Proceed to the next layer of feature identification.

[0112] Third layer: Euclidean distance matching and filtering based on spectral features

[0113] Construct a 5-dimensional normalized spectral feature vector: This includes the normalized peak position, 3dB normalized bandwidth, spectral roll-off factor, sidelobe attenuation, and time-frequency energy entropy; it also calculates the Euclidean distance between the measured feature and the clutter standard feature library. Set distance threshold If the distance is less than the threshold, it is judged as non-target clutter and discarded; otherwise, the suspicious drone target signal is retained and proceeds to the next layer of feature identification.

[0114] Fourth layer: System identification based on known signal correlation coefficients

[0115] The similarity between the current signal and the known UAV spectrum signal is calculated using the Pearson correlation coefficient. The calculation formula is as follows: Preset relevant thresholds , Signals identified as already documented or known will be directly removed. The signal was determined to be from an unknown drone, and the process proceeded to the next step.

[0116] Fifth layer: Multidimensional hash deduplication filtering

[0117] Combined center frequency ,bandwidth timestamp Multidimensional feature vectors Generate hash value: The signal is compared with the local hash index table. If a match is found, the signal is considered a duplicate and is removed. If no match is found, the signal is confirmed as a new, unfamiliar, and valid drone signal.

[0118] Step 3: Multidimensional feature extraction and normalized metadata encapsulation;

[0119] For signals that have passed the multi-dimensional feature joint screening, the modulation mode, symbol rate, frequency hopping period, and intra-pulse coding multi-dimensional features are extracted; the sampling rate, data bit width, and IQ storage format are unified; and the acquisition time, latitude and longitude, azimuth angle, and airspace scene metadata are bound to form a standardized UAV spectrum fingerprint data packet.

[0120] For signals that have passed the multi-dimensional feature joint screening, the following multi-dimensional feature parameters are extracted:

[0121] Time-frequency characteristics: characteristics of signal time-frequency distribution diagram, characteristics of instantaneous frequency change curve;

[0122] Modulation characteristics: modulation scheme (such as ASK, FSK, PSK, QAM, etc.), symbol rate, frequency hopping period, frequency hopping sequence;

[0123] Intrapulse characteristics: intrapulse coding method (such as Barker code, P4 code, Frank code, etc.), intrapulse modulation parameters.

[0124] Perform data format normalization on the above multidimensional parameters: unify the sampling rate to the standard sampling rate, unify the data bit width to the standard bit width (such as 16-bit IQ), and unify the IQ storage format (such as binary interleaved storage).

[0125] Simultaneously, the collected metadata is bound: collection time (UTC time format), GPS latitude and longitude coordinates, azimuth angle (angle relative to due north), and airspace scene labels (such as urban areas, border ports, key locations, etc.), forming a standardized UAV spectrum fingerprint data packet.

[0126] Step 4: On-site real-time warehousing and subsequent signal sorting and identification;

[0127] Standardized unfamiliar UAV spectrum fingerprint data packets are written into the spectrum feature database on-site to establish a joint index of frequency points and multi-dimensional features. Specifically, a joint index structure is constructed with the center frequency as the primary key and feature parameters such as modulation scheme, symbol rate, and frequency hopping period as secondary keys.

[0128] When a drone of the same type re-enters the country, the Euclidean distance of the third layer and the correlation coefficient of the fourth layer are reused to complete multi-dimensional feature matching: First, the candidate template set is quickly located by frequency indexing, and then the Euclidean distance and correlation coefficient between the signal to be tested and the template in the library are calculated one by one. When the Euclidean distance between the selected template and the signal to be tested is less than the preset threshold and the correlation coefficient is greater than the preset threshold, it is determined to be a successful match, realizing automatic sorting, identity recognition, target trajectory association and low-altitude situational alarm.

[0129] This invention constructs a complete closed-loop process for multi-dimensional feature joint identification and self-iterative data storage through the above steps. The parameters of each layer in the five-layer quantitative screening system can be adaptively adjusted and optimized according to the actual electromagnetic environment and detection requirements, exhibiting high engineering flexibility and environmental adaptability.

Claims

1. A method for detecting and storing UAV signals based on multi-dimensional feature joint identification, characterized in that, Includes the following steps: Step S1, Acquisition of raw electromagnetic spectrum signals: Broadband spectrum interception is performed on the monitored airspace through a broadband electronic reconnaissance receiver, and raw electromagnetic spectrum signals are acquired in real time. The output includes time-domain discrete sequence, frequency-domain FFT sequence, IQ sampling data and basic signal characteristic parameters, and the raw spectrum dataset is constructed. Step S2, Multi-dimensional feature joint identification and five-layer quantization screening: Construct a multi-dimensional feature joint identification and five-layer quantization algorithm screening system, which sequentially uses adaptive noise floor estimation algorithm, signal-to-noise ratio quantization decision algorithm, spectrum feature Euclidean distance matching algorithm, known signal library feature correlation coefficient identification algorithm, and multi-dimensional hash deduplication algorithm to eliminate environmental noise floor, clutter interference, known communication radar signals, archived UAV signals and duplicate collected signals, retaining only the effective radiation signals of unknown UAVs that have not been archived; Step S3, Multi-dimensional feature extraction and metadata encapsulation: For unfamiliar UAV signals that have passed the joint screening of multi-dimensional features, extract multi-dimensional parameters such as time-frequency features, modulation features, and intra-pulse features, and complete the data format normalization and integrated encapsulation of the acquisition time, space, orientation, and scene metadata. Step S4, on-site real-time storage and subsequent signal sorting and identification: The packaged unfamiliar UAV spectrum feature fingerprint is written into the local spectrum feature library on-site, and a new exclusive template index entry is added; when the unfamiliar UAV of the same type reappears in the reconnaissance airspace, multi-dimensional feature matching and sorting are completed based on the template in the library to achieve accurate target identification and situation warning.

2. The method for detecting and storing UAV signals based on multi-dimensional feature joint identification according to claim 1, characterized in that, In step S1, a discrete-time signal sequence is obtained by continuously sampling the monitored spatial domain through a broadband radio frequency receiving front-end. , The frequency domain spectrum sequence is obtained by FFT transformation. ; The fundamental characteristic parameters of a signal include its center frequency. Bandwidth usage The parameters include signal arrival time (TOA), signal pulse width (PW), signal amplitude, time-domain energy, and spectral envelope shape.

3. The method for detecting and storing UAV signals based on multi-dimensional feature joint identification according to claim 1, characterized in that, In step S2, the first-layer adaptive noise floor estimation algorithm uses a sliding window recursive mean model to estimate the noise power, calculated according to the following formula: in, For the first Point real-time noise power estimate, For smooth forgetting factor, The amplitude of the signal in the time domain; Set the noise floor threshold ,in This is the noise margin coefficient, when If the noise is determined to be environmental background noise, it will be directly removed; otherwise, it will be retained and proceed to the next level of multi-dimensional feature identification.

4. The method for detecting and storing UAV signals based on multi-dimensional feature joint identification according to claim 3, characterized in that, In step S2, the signal-to-noise ratio (SNR) quantization decision algorithm of the second layer calculates the SNR according to the following formula. : in The effective frequency band of the signal; This is the pure noise frequency band. N and M are the number of signal and noise sampling points, respectively; For the first The spectral amplitude value of each frequency point; Set low threshold High threshold ;: when Invalid weak signals are directly removed. hour The time stamp is questionable and needs further verification; when Then proceed to the next layer of multi-dimensional feature joint identification.

5. The method for detecting and storing UAV signals based on multi-dimensional feature joint discrimination according to claim 4, characterized in that, In step S2, the Euclidean distance matching algorithm for the spectral features of the third layer is executed as follows: Constructing a 5-dimensional normalized spectral feature vector ,in For the normalized position of the spectral peak, For 3dB normalized bandwidth, For the spectral roll-off factor, For sidelobe attenuation, The time-frequency energy entropy; Calculate the Euclidean distance between the eigenvector of the signal under test and the standard eigenvector of known clutter in the library: In the formula, The feature vector of the signal to be measured. The standard eigenvectors of known clutter within the library; Set distance threshold ,like If it is determined to be clutter, it will be removed; otherwise, it will be retained as a suspicious drone target signal and allowed to proceed to the next layer.

6. The method for detecting and storing UAV signals based on multi-dimensional feature joint discrimination according to claim 5, characterized in that, In step S2, the fourth layer's known signal library feature correlation coefficient discrimination algorithm uses the Pearson correlation coefficient. Calculate the similarity between the current signal and known signals in the database using the following formula: In the formula, This is the current signal feature vector. The feature vectors of known signals within the library; Set the relevant threshold as ,when Signals that are known to have been archived are directly removed; only those that are retained are removed. The system identifies unfamiliar and unknown drone signals and performs system-level feature identification.

7. The method for detecting and storing UAV signals based on multi-dimensional feature joint discrimination according to claim 6, characterized in that, In step S2, the multidimensional hash deduplication algorithm of the fifth layer is executed as follows: Combined center frequency Bandwidth usage Collection timestamp eigenvectors Jointly generate hash value The generated hash value is compared with the local hash table. If they match, the signal is considered a duplicate and is removed. If they do not match, the signal is confirmed as a new, unfamiliar, and valid drone signal.

8. The method for detecting and storing UAV signals based on multi-dimensional feature joint identification according to claim 1, characterized in that, In step S3, the extracted time-frequency features include signal time-frequency distribution map features and instantaneous frequency change features; the extracted modulation features include modulation scheme, symbol rate, frequency hopping period, and frequency hopping sequence; the extracted intra-pulse features include intra-pulse coding scheme and intra-pulse modulation parameters. Data format normalization includes uniform sampling rate, uniform data bit width, and uniform IQ storage format; Metadata includes the collection time (UTC time), GPS latitude and longitude coordinates, azimuth angle, and airspace scene label.

9. A UAV signal detection and database entry system based on multi-dimensional feature joint discrimination, used to implement the UAV signal detection and database entry method based on multi-dimensional feature joint discrimination as described in any one of claims 1 to 8, characterized in that, include: The signal acquisition module is used to intercept the broadband spectrum of the monitored airspace through the broadband radio frequency receiving front end, acquire the original electromagnetic spectrum signal on site in real time, and output the time-domain discrete sequence, frequency-domain FFT sequence, IQ sampling data and basic signal characteristic parameters. The multi-level joint screening module is used to perform multi-dimensional feature joint screening and five-level quantization algorithm screening. This module includes an adaptive noise floor estimation sub-module, a signal-to-noise ratio quantization decision sub-module, a spectrum feature Euclidean distance matching sub-module, a known signal library feature correlation coefficient screening sub-module, and a multi-dimensional hash deduplication sub-module connected in sequence. The feature extraction and encapsulation module is used to extract time-frequency features, modulation features, intra-pulse features and multi-dimensional parameters from the filtered unfamiliar drone signals, perform data format normalization and bind the acquired spatiotemporal, azimuth and scene metadata; The spectrum feature library entry module is used to write the packaged unfamiliar UAV spectrum feature fingerprint into the local spectrum feature library and create index entries. The signal matching and recognition module is used to perform multi-dimensional feature matching and sorting based on templates in the library when reproducing similar UAVs in the future, so as to achieve target recognition and situational warning.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the UAV signal detection and database entry method based on multi-dimensional feature joint discrimination as described in any one of claims 1 to 8.

Images