CN110097011A - A kind of signal recognition method and device - Google Patents
A kind of signal recognition method and device Download PDFInfo
- Publication number
- CN110097011A CN110097011A CN201910371624.7A CN201910371624A CN110097011A CN 110097011 A CN110097011 A CN 110097011A CN 201910371624 A CN201910371624 A CN 201910371624A CN 110097011 A CN110097011 A CN 110097011A
- Authority
- CN
- China
- Prior art keywords
- signal
- frequency image
- sample
- target time
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention provides a kind of signal recognition method and devices.Scheme is as follows: the signal to be identified in available default frequency range, determine the target time-frequency image of signal to be identified, extract the target signature data of target time-frequency image, target signature data are inputted into trained signal identification model in advance, determine the type of signal to be identified, signal identification model is the model obtained by presetting training set training, default training set includes the sample type of the sample characteristics data of the corresponding sample time-frequency image of multiple sample signals and each sample signal in default frequency range.The technical solution provided through the embodiment of the present invention, according to the difference of the characteristic in the corresponding time-frequency image of unlike signal, utilize trained signal identification model, determine the type of signal, it realizes and identifies different types of signal using an equipment, it no longer needs individually to be equipped with corresponding module for the signal of each type, reduces the cost of equipment, improve signal identification efficiency.
Description
Technical field
The present invention relates to signal detection technical fields, more particularly to a kind of signal recognition method and device.
Background technique
In conventional navigation position fixing process, often through identification Global Navigation Satellite System (Global Navigation
Satellite System, GNSS) signal positioned.With the continuous development of technology, there are more localization methods, such as
Full source navigation, opportunity signal navigation, multi-source fusion positioning etc..Using these positioning methods, often by institute in identification zone
There is the radiofrequency signal that can be used for positioning to be positioned.The radiofrequency signal may include various non-navigational special signals, word tone in full
Frequency broadcast, digital TV broadcast signal, amplitude modulation and fm broadcast signal, cellular base station signal, bluetooth (Bluetooth) signal,
Purple honeybee (ZigBee) signal, wireless network (Wi-Fi) signal etc..Unlike GNSS signal, these radiofrequency signals are distributed in
In wider frequency range, and modulation system used by each signal is different, this brings larger for signal identification process
Difficulty.
Currently, identifying each communication module by different communication modules when identifying to above-mentioned radiofrequency signal
Corresponding radiofrequency signal, this makes equipment cost higher, and signal identification efficiency is lower.
Summary of the invention
The embodiment of the present invention be designed to provide a kind of signal recognition method and device is mentioned with reducing the cost of equipment
High RST recognition efficiency.Specific technical solution is as follows:
The embodiment of the invention provides a kind of signal recognition methods, comprising:
Obtain the signal to be identified in default frequency range;
Determine the target time-frequency image of the signal to be identified;
Extract the target signature data of the target time-frequency image;
The target signature data are inputted into trained signal identification model in advance, determine the class of the signal to be identified
Type, wherein the signal identification model is the model obtained by presetting training set training, and the default training set includes described
The sample of the sample characteristics data of the corresponding sample time-frequency image of multiple sample signals and each sample signal in default frequency range
Type.
Optionally, the step of target time-frequency image of the determination signal to be identified, comprising:
Using following Short Time Fourier Transform formula, the target time-frequency image of the signal to be identified is obtained:
Gf(w, u)=∫ f (t) g (t-u) e-jwtdt
Wherein, w is the angular frequency of the signal to be identified, and f is frequency, and t is time t, and u is preset time length of window u,
Function GfThe value of (w, u) is the amplitude of frequency component, and ∫ dt is the integration operation to t, and function f (t) is the letter to be identified
Number, function g (t-u) is preset window function, function e-jwtFor complex function, e is natural constant, and j is imaginary unit.
Optionally, the step of target signature data extracted in the target time-frequency image, comprising:
Feature extraction is carried out to the target time-frequency image, obtains multiple characteristic points of the target time-frequency image;
Using K mean cluster (K-means) algorithm, clustering processing is carried out to the multiple characteristic point, obtains K class;
Determine the quantity for the characteristic point that each class includes in K class;
According to the quantity for the characteristic point that each class includes, the target signature data in the target time-frequency image are determined.
Optionally, described that feature extraction is carried out to the target time-frequency image, obtain the multiple of the target time-frequency image
The step of characteristic point, comprising:
Using acceleration robust feature (Speeded-Up Robust Features, SURF) algorithm, when extracting the target
Multiple feature descriptors in frequency image obtain multiple characteristic points of the target time-frequency image.
Optionally, the quantity of the characteristic point for including according to each class determines the target in the target time-frequency image
The step of characteristic, comprising:
According to the quantity for the characteristic point that each class includes, the fisrt feature point distribution map of the target time-frequency image is constructed;
The target signature data fisrt feature point distribution map being determined as in the target time-frequency image;Or,
According to the quantity for the characteristic point that each class includes, characteristic point that each class includes is counted in the target time-frequency image
The probability of middle appearance, and according to the corresponding probability of each class, construct the second feature point distribution map of the target time-frequency image;It will
The second feature point distribution map is determined as the target signature data in the target time-frequency image.
Optionally, the signal identification model is obtained using following steps training, comprising:
Obtain the default training set;
Multiple sample characteristics data are inputted into preset machine learning model respectively, determine the type of each sample signal;
According to the sample type of the type of determining each sample signal and each sample signal, penalty values are calculated;
Judge whether penalty values are less than preset threshold;
If it is not, then adjusting the parameter of preset machine learning model, it is described by multiple sample characteristics data point to return to execution
The step of not inputting preset machine learning model, determining the type of each sample signal;
If so, preset machine learning model is determined as signal identification model.
The embodiment of the invention also provides a kind of signal recognition devices, comprising:
First obtains module, for obtaining the signal to be identified in default frequency range;
First determining module, for determining the target time-frequency image of the signal to be identified;
Extraction module, for extracting the target signature data of the target time-frequency image;
Second determining module is determined for the target signature data to be inputted trained signal identification model in advance
The type of the signal to be identified, wherein the signal identification model is the model obtained by presetting training set training, described
Default training set includes the sample characteristics data of the corresponding sample time-frequency image of multiple sample signals in the default frequency range, and
The sample type of each sample signal.
Optionally, first determining module is specifically used for utilizing following Short Time Fourier Transform formula, obtain it is described to
The target time-frequency image of identification signal:
Gf(w, u)=∫ f (t) g (t-u) e-jwtdt
Wherein, w is the angular frequency of the signal to be identified, and f is frequency, and t is time t, and u is preset time length of window u,
Function GfThe value of (w, u) is the amplitude of frequency component, and ∫ dt is the integration operation to t, and function f (t) is the letter to be identified
Number, function g (t-u) is preset window function, function e-jwtFor complex function, e is natural constant, and j is imaginary unit.
Optionally, the extraction module, comprising:
Extracting sub-module obtains the target time-frequency image for carrying out feature extraction to the target time-frequency image
Multiple characteristic points;
Submodule is clustered, for using K-means clustering algorithm, clustering processing is carried out to the multiple characteristic point, obtains K
A class;
First determines submodule, for determining the quantity of each class includes in K class characteristic point;
Second determines submodule, and the quantity of the characteristic point for including according to each class determines the target time-frequency image
In target signature data.
Optionally, the extracting sub-module is specifically used for utilizing SURF algorithm, extract more in the target time-frequency image
A feature descriptor obtains multiple characteristic points of the target time-frequency image.
Optionally, it described second determines submodule, specifically for the quantity for the characteristic point for including according to each class, constructs institute
State the fisrt feature point distribution map of target time-frequency image;The fisrt feature point distribution map is determined as the target time-frequency image
In target signature data;Or the quantity for the characteristic point for according to each class including, characteristic point that each class includes is counted described
The probability occurred in target time-frequency image, and according to the corresponding probability of each class, construct the target time-frequency image second is special
Sign point distribution map;The target signature data second feature point distribution map being determined as in the target time-frequency image.
Optionally, described device further include:
Second obtains module, for obtaining the default training set;
Third determining module determines every for multiple sample characteristics data to be inputted preset machine learning model respectively
The type of one sample signal;
Computing module, for according to the type of determining each sample signal and the sample type of each sample signal, meter
Calculate penalty values;
Judgment module, for judging whether penalty values are less than preset threshold;
Adjust module, for the judging result in the judgment module be it is no when, adjust preset machine learning model
Parameter, return execution is described to input preset machine learning model for multiple sample characteristics data respectively, determines each sample letter
Number type the step of;
4th determining module, for the judgement of the judgment module be result be when, by preset machine learning mould
Type is determined as signal identification model.
The embodiment of the invention also provides a kind of electronic equipment, including processor, communication interface, memory and communication are total
Line, wherein processor, communication interface, memory complete mutual communication by communication bus;
Memory, for storing computer program;
Processor when for executing the program stored on memory, realizes any of the above-described signal recognition method
Step.
The embodiment of the invention also provides a kind of computer readable storage medium, the computer readable storage medium memory
Computer program is contained, the computer program realizes any of the above-described signal recognition method step when being executed by processor
Suddenly.
The embodiment of the invention also provides a kind of computer program products comprising instruction, when it runs on computers
When, so that computer executes any of the above-described signal recognition method.
A kind of signal recognition method and device provided in an embodiment of the present invention, the letter to be identified in available default frequency range
Number, it determines the target time-frequency image of signal to be identified, the target signature data of target time-frequency image is extracted, by target signature data
Trained signal identification model in advance is inputted, determines the type of signal to be identified, wherein signal identification model is by default
The obtained model of training set training, default training set include the corresponding sample time-frequency image of multiple sample signals in default frequency range
The sample type of sample characteristics data and each sample signal.The technical solution provided through the embodiment of the present invention, according to not
The difference of characteristic with signal in corresponding time-frequency image determines the class of signal using trained signal identification model
Type is realized and identifies different types of signal using an equipment, it is no longer necessary to individually be equipped with correspondence for the signal of each type
Module, reduce the cost of equipment, improve signal identification efficiency.
Certainly, implement any of the products of the present invention or method it is not absolutely required at the same reach all the above excellent
Point.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of flow diagram of signal recognition method provided in an embodiment of the present invention;
Fig. 2-a is a kind of time-frequency figure provided in an embodiment of the present invention carried out after Short Time Fourier Transform to Wi-Fi signal
Picture;
Fig. 2-b is a kind of time-frequency figure provided in an embodiment of the present invention carried out after Short Time Fourier Transform to Bluetooth signal
Picture;
Fig. 2-c is a kind of time-frequency figure provided in an embodiment of the present invention carried out after Short Time Fourier Transform to ZigBee signal
Picture;
Fig. 3-a is a kind of schematic diagram of the scale space of target time-frequency image provided in an embodiment of the present invention;
Fig. 3-b is a kind of schematic diagram that the principal direction of characteristic point provided in an embodiment of the present invention determines;
Fig. 4 is a kind of flow diagram of the training method of signal identification model provided in an embodiment of the present invention;
Fig. 5-a is that sample signal provided in an embodiment of the present invention is one of corresponding sample time-frequency image of Bluetooth signal;
Fig. 5-b is that sample signal provided in an embodiment of the present invention is one of corresponding sample time-frequency image of Bluetooth signal;
Fig. 5-c is that sample signal provided in an embodiment of the present invention is one of corresponding sample time-frequency image of Bluetooth signal;
Fig. 6 is a kind of structural schematic diagram of signal recognition device provided in an embodiment of the present invention;
Fig. 7 is a kind of structural schematic diagram of electronic equipment provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Currently, when being identified to radiofrequency signal, according to communication protocol used by different types of radiofrequency signal
Difference utilizes the radiofrequency signal of different communication module identification corresponding types.It is unfavorable since different traffic models is mutually indepedent
In the integrated and depth integration of equipment, so that the higher cost of the equipment of identification signal type.In addition, each communication module is only capable of
The signal for identifying the communication module corresponding types causes signal identification efficiency lower.
In order to solve above equipment higher cost and the lower problem of signal identification efficiency, the embodiment of the present invention is provided
A kind of signal recognition method.This method is applied to any electronic equipment including navigation system or positioning system.In the present invention
In the method that embodiment provides, the signal to be identified in available default frequency range determines the target time-frequency figure of signal to be identified
Picture extracts the target signature data of target time-frequency image, and target signature data are inputted trained signal identification model in advance,
Determine the type of signal to be identified, wherein signal identification model is the model obtained by presetting training set training, presets training
Sample characteristics data and each sample signal of the collection including the corresponding sample time-frequency image of sample signals multiple in default frequency range
Sample type.
The technical solution provided through the embodiment of the present invention, according to the characteristic in the corresponding time-frequency image of unlike signal
Difference determine the type of signal using trained signal identification model, realize and identify different type using an equipment
Signal, it is no longer necessary to be individually equipped with corresponding module for the signal of each type, reduce the cost of equipment, improve signal
Recognition efficiency.
Below by specific embodiment, the embodiment of the present invention is illustrated.
As shown in FIG. 1, FIG. 1 is a kind of flow diagrams of signal recognition method provided in an embodiment of the present invention.This method
Include the following steps.
Step S101 obtains the signal to be identified in default frequency range.
In this step, electronic equipment obtains the radiofrequency signal in default frequency range, as signal to be identified.
Above-mentioned radiofrequency signal may include Bluetooth signal, Wi-Fi signal and ZigBee signal etc., and above-mentioned default frequency range can be with
It, can also be with for 2.4 Gigahertzs (GHz) industrial scientific medical (Industrial Scientific Medical, ISM) frequency range
For 5.0GHz frequency range.For example, default frequency range can be 2.4GHz frequency range.A certain moment, the available 2.4GHz frequency of electronic equipment
Radiofrequency signal in section, as signal to be identified.
Above-mentioned default frequency range can be set according to practical application scene and the demand of user etc..For example, using wide
When broadcasting signal and being navigated or positioned, broadcast singal may include digital audio broadcasting signal, digital TV broadcast signal, frequency modulation
Broadcast singal and amplitude modulated broadcast signal etc., these broadcast singals are distributed in different frequency ranges, then above-mentioned default frequency range can be with
For frequency range corresponding to each broadcast singal.In embodiments of the present invention, to the quantity of above-mentioned default frequency range and default frequency range
It is not especially limited.
Step S102 determines the target time-frequency image of signal to be identified.
In this step, signal to be identified accessed by above-mentioned steps S101 is time-domain signal, and electronic equipment can be right
Signal to be identified carries out certain conversion process, that is, time-domain signal is converted to frequency-region signal, to obtain letter to be identified
Number time-frequency image, as target identification image.
In an optional embodiment, the step of the target time-frequency image of above-mentioned determination signal to be identified, it can indicate are as follows:
Using following Short Time Fourier Transform formula, obtain the target time-frequency image of signal to be identified, that is, utilize with
Lower formula treats identification signal and carries out Short Time Fourier Transform, obtains the target time-frequency image of signal to be identified.Wherein, in short-term
Fourier formula can indicate are as follows:
Gf(w, u)=∫ f (t) g (t-u) e-jwtdt
W is the angular frequency of signal to be identified, and f is frequency, and t is time t, and u is preset time length of window u, function Gf(w,
U) value is the amplitude of frequency component, and ∫ dt is the integration operation to t, and function f (t) is signal to be identified, and function g (t-u) is
Preset window function, function e-jwtFor complex function, e is natural constant, and j is imaginary unit.
In embodiments of the present invention, since above-mentioned signal to be identified is often nonstationary random signal, if directly treating knowledge
Level signal carries out Fourier transformation, and certain error will be present in the result after Fourier transformation.When in order to improve above-mentioned target
The accuracy of frequency image treats identification signal using preset window function and carries out Short Time Fourier Transform.Namely will be wait know
When level signal is converted to frequency-region signal by time-domain signal, by lesser time window, the interception time window in signal to be identified
Signal in mouthful, and to the signal being truncated into Fourier transformation.Since the length of preset time window is smaller, so that the time
The signal being truncated in window is considered stationary random signal, moves the time window in Signal Move to be identified, completes
Treat the Short Time Fourier Transform process of identification signal, and then available target time-frequency image.
By taking Wi-Fi signal, Bluetooth signal and ZigBee signal as an example.To Wi-Fi signal, Bluetooth signal and
After ZigBee signal carries out Short Time Fourier Transform, available time-frequency image as shown in Figure 2.Wherein, Fig. 2-a is the present invention
A kind of time-frequency image Wi-Fi signal carried out after Short Time Fourier Transform that embodiment provides.Fig. 2-b is the embodiment of the present invention
What is provided carries out a kind of time-frequency image after Short Time Fourier Transform to Bluetooth signal.Fig. 2-c is provided in an embodiment of the present invention
A kind of time-frequency image after carrying out Short Time Fourier Transform to ZigBee signal.In Fig. 2-a, Fig. 2-b and Fig. 2-c, level side
To the frequency for indicating signal, vertical direction indicates the time, and the black region in time-frequency image is multiple sampled points composition, each to adopt
Sampling point is expressed as the amplitude of frequency component at corresponding sampled point.
In an optional embodiment, above-mentioned preset window function can according to the actual situation and the selection such as user demand
Different window functions, such as Hanning window (Hanning), Hamming window (Hamming), Blackman window (Blackman).
In an optional embodiment, the preset time length of window of above-mentioned preset window function can be according in time-frequency figure
Temporal resolution and frequency resolution determine.For example, being cut in preset time window when preset time length of window is longer
The signal to be identified got is also longer, and the frequency resolution after Short Time Fourier Transform is higher, and temporal resolution is lower.When pre-
If time window length is more in short-term, the signal to be identified being truncated in preset time window is also shorter, and Fourier becomes in short-term
Frequency resolution after changing is lower, and temporal resolution is higher.
In embodiments of the present invention, above-mentioned preset window function and preset time window are not especially limited.
In an optional embodiment, before the target time-frequency image for determining above-mentioned signal to be identified, that is, treat
Before identification signal carries out Short Time Fourier Transform, electronic equipment can treat identification signal and carry out certain pretreatment.For example,
It treats identification signal to be filtered, filters out the noise signal etc. in signal to be identified.In embodiments of the present invention, to pre- place
Reason process and mode are not especially limited.
Step S103 extracts the target signature data of target time-frequency image.
In this step, electronic equipment can carry out feature extraction to the target time-frequency image of above-mentioned signal to be identified, obtain
To target signature data.
In an optional embodiment, can use preset algorithm or default network model to above-mentioned signal to be identified into
Row feature extraction obtains the target signature data of target time-frequency image.Here, not making to preset algorithm and default network model
It illustrates.
Target signature data are inputted trained signal identification model in advance, determine signal to be identified by step S104
Type.
In this step, by the target signature data input of the target time-frequency image extracted, trained signal is known in advance
In other model.According to the target signature data and the structure and parameter of the signal identification model, available signal to be identified
Type.Wherein, signal identification model is the model obtained by presetting training set training.Default training set may include presetting
The sample characteristics data of the corresponding sample time-frequency image of multiple sample signals and the sample class of each sample signal in frequency range
Type.
In one embodiment, above-mentioned signal identification model, can be according to extracting when determining the type of signal to be identified
Signal to be identified target signature data and sample signal in default training set sample characteristics data, determine target spy
Levy the size of the degree of conformity between data and the sample characteristics data of different sample signals.It, will be with mesh according to determining degree of conformity
The sample class of the corresponding sample signal of the mark maximum sample characteristics data of characteristic degree of conformity is determined as signal to be identified
Type.Wherein, degree of conformity can be indicated with mahalanobis distance, Euclidean distance etc., by taking Euclidean distance as an example, if Euclidean distance is smaller,
Then above-mentioned degree of conformity is bigger.If Euclidean distance is bigger, above-mentioned degree of conformity is smaller.
Still it is illustrated by taking above-mentioned Wi-Fi signal, Bluetooth signal and ZigBee signal as an example.If above-mentioned signal identification model
For the identification model for these three signals, then above-mentioned sample signal is Wi-Fi signal, Bluetooth signal and ZigBee signal, on
Sample characteristics data are stated as the characteristic in sample time-frequency image corresponding to Wi-Fi signal, Bluetooth signal and ZigBee signal
According to.
In an optional embodiment, for above-mentioned steps S103, the target signature data of target time-frequency image, tool are extracted
Body may comprise steps of.
Step S1031 carries out feature extraction to target time-frequency image, obtains multiple characteristic points of target time-frequency image.
In this step, electronic equipment can use above-mentioned preset algorithm or default network model extracts signal to be identified
Multiple characteristic points in target time-frequency image obtain multiple characteristic points of target time-frequency image.
Step S1032 is carried out clustering processing to multiple characteristic points, is obtained K class using K-means algorithm.
In this step, electronic equipment can use multiple spies of the K-means algorithm to the target time-frequency image extracted
Sign point carries out clustering processing, obtains K class.
Step S1033 determines the quantity for the characteristic point that each class includes in K class.
In this step, for each of above-mentioned K class class, the number for the characteristic point for including in such can be counted
Amount, so that it is determined that the quantity for the characteristic point for including in each class in K class.
Step S1034 determines the target signature number in target time-frequency image according to the quantity for the characteristic point that each class includes
According to.
In an optional embodiment, according to the quantity for the characteristic point that each class includes, the of target time-frequency image is constructed
One characteristic point distribution map;Fisrt feature point distribution map is determined as the target signature data in target time-frequency image.
Specifically, electronic equipment can be according to the quantity and each characteristic point for the characteristic point for including in each class upper
State the position in target time-frequency image, the distribution map of construction feature point, the fisrt feature point distribution map as target time-frequency image.
According to fisrt feature point distribution map, electronic equipment can determine the target signature data of above-mentioned target time-frequency image, when such as target
The quantity for the characteristic point for including in the distribution situation of characteristic point in frequency image, the distribution situation of each class and each class.
In another optional embodiment, according to the quantity for the characteristic point that each class includes, the spy that each class includes is counted
The probability that sign point occurs in target time-frequency image, and according to the corresponding probability of each class, construct the second of target time-frequency image
Characteristic point distribution map;Second feature point distribution map is determined as the target signature data in target time-frequency image.
Specifically, electronic equipment can count in each class according to the quantity for the characteristic point for including in each class comprising spy
The probability that sign point occurs in target time-frequency image, and according to the corresponding probability of each class, construction feature point distribution map, as mesh
Mark the second feature point distribution map of time-frequency image.Electronic equipment can determine the target signature data of above-mentioned target time-frequency image,
Such as the distribution situation of characteristic point in target time-frequency image, the characteristic point that includes in the distribution situation of each class and each class
Quantity, the probability etc. that the characteristic point that each class includes occurs.
By carrying out feature extraction to target time-frequency image, and clustering processing is carried out to the multiple characteristic points extracted, obtained
To target signature data, the accuracy of target signature data is improved, so that improving signal identification model determines letter to be identified
Number classification accuracy, improve signal identification efficiency.
In an optional embodiment, in above-mentioned steps S1031, feature extraction is carried out to target time-frequency image, obtains mesh
During the multiple characteristic points for marking time-frequency image, electronic equipment can use SURF algorithm, extract in target time-frequency image
Multiple feature descriptors obtain multiple characteristic points of target time-frequency image.It can specifically include following steps:
Step S1031A determines the corresponding integral image of target time-frequency image.
In this step, electronic equipment can carry out Integral Processing to above-mentioned target time-frequency image, obtain target time-frequency figure
As corresponding integral image.
In the integral image, if using the pixel of the left upper of the integral image as coordinate origin, horizontal direction right
To for X-direction, vertically downward direction is Y direction, then it is the pixel that coordinate value, which is the integrated pixel value of pixel at (x, y),
The sum of point pixel value of all pixels point into the rectangular area in the integral image upper left corner, that is, the pixel and coordinate original
Point constitutes the sum of the pixel value of all pixels point in rectangular area.It can specifically indicate are as follows:
Wherein, I (x, y) is the integrated pixel value of pixel (x, y), I (xi,yj) it is pixel (xi,yj) pixel value.
Step S1031B carries out process of convolution to integral image, obtains the corresponding scale space of target time-frequency image.Its
In, scale space is the form of expression in image under different resolutions.
In an optional embodiment, electronic equipment can use tank filters and handle above-mentioned integral image, that is,
Constantly change the size of tank filters, and by the tank filters and integral image progress convolution after change, when obtaining target
The gaussian pyramid scale space of frequency image, the scale space as above-mentioned target time-frequency image.
When constructing above-mentioned gaussian pyramid scale space, that is, building target time-frequency image scale space when,
SURF algorithm uses black plug (Hessian) matrix determinant approximation image, then the integrated pixel value of pixel (x, y)
Hessian matrix H (I (x, y)) can indicate are as follows:
Wherein,WithThe knot for the second order local derviation for being integrated pixel value I at pixel (x, y)
Fruit, I are the integrated pixel value at pixel (x, y).
SURF algorithm is using second order standard gaussian function g (x, y, σ) to above-mentioned Hessian matrix determinant approximation image
Process of convolution is carried out, gaussian pyramid scale space, that is, the scale space of building target time-frequency image are constructed.At this point, right
In above-mentioned pixel (x, y), in the case where scale is σ, above-mentioned Hessian matrix can be indicated are as follows:
Wherein, Lxx(x,y,σ)、Lxy(x,y,σ)、Lyy(x, y, σ) is the second order canonical function g in the case where scale is σ
The result of second order local derviation convolution at pixel (x, y) of (x, y, σ).
Step S1031C determines the position of multiple characteristic points in scale space, obtains multiple characteristic points.
In this step, the position of the local maximum corresponding pixel points of above-mentioned H (x, y, σ) is chosen in scale space
It sets, as the position of characteristic point, obtains multiple characteristic points.
It is illustrated by taking Fig. 3-a as an example, Fig. 3-a is the scale space of target time-frequency image provided in an embodiment of the present invention
A kind of schematic diagram.Wherein, it is σ that image 301, which is scale,iWhen target time-frequency image and above-mentioned second order standard gaussian convolution of functions after
Image, pixel 302 are the pixel in image 301.Being respectively present scale in the top of image 301 and lower section is σi+1
And σi-1Image (not showed that in Fig. 3-a) after corresponding target time-frequency image and second order standard gaussian convolution of functions.In determination
When characteristic point in image 302, when such as determining whether pixel 302 is characterized, it is thus necessary to determine that H at pixel 302 (x, y,
It whether is σ) its 26 adjacent pixel (at most 8 pixels adjacent with pixel 302 in image 301, with image 301
At most 18 pixels adjacent with pixel 302 in adjacent image) in it is maximum.If the H's (x, y, σ) of pixel 302
Value is the largest, then can determine that pixel 302 is characterized a little.It, then can be with if the value of pixel H (x, y, σ) is not the largest
Determine that pixel 302 is not characteristic point.
Step S1031D determines the principal direction of each characteristic point.
In this step, in order to improve SURF algorithm to rotationally-varying robustness, that is, in order to improve target time-frequency
The robustness for the characteristic point that image extracts after rotating can determine the principal direction of each characteristic point.
It for convenience of understanding, is illustrated by taking Fig. 3-b as an example, Fig. 3-b is the main side of characteristic point provided in an embodiment of the present invention
To a kind of determining schematic diagram.In the border circular areas using characteristic point as the center of circle, i.e., in region 303, according to default rotation angle,
Such as α, the rotating fan region 304 in region 303 calculates all characteristic points in fan-shaped region 304 in the horizontal direction and vertically
The response of Ha Er (Haar) small echo on direction, the maximum fan-shaped direction of Haar small echo accumulated value in fan-shaped region 304 is true
It is set to the principal direction of characteristic point.Wherein, the angle of fan-shaped region is predetermined angle, such as 60 °.In embodiments of the present invention, to upper
The calculation method for stating the response of Haar small echo does not specifically illustrate.
Step S1031E determines the corresponding feature of this feature point according to the principal direction of this feature point for each characteristic point
Vector, as feature descriptor.
In this step, the rectangle region of default size is chosen according to the principal direction of this feature point for each characteristic point
Domain, and the rectangular area is divided into multiple subregions.Count the pixel of preset quantity in each subregion in the horizontal direction
With the response of Haar small echo in vertical direction.According to the sum of the response of horizontal direction Haar small echo, horizontal direction Haar is small
The sum of the absolute value of the response of wave, the sum of the response of vertical direction Haar small echo and the sound of vertical direction Haar small echo
The sum for the absolute value that should be worth generates feature vector, the feature descriptor as this feature point.Wherein, horizontal direction and Vertical Square
To for horizontally and vertically with respect to this feature point principal direction.
The feature descriptor of features described above pointIt can indicate are as follows:
Wherein, X is the horizontal direction of the principal direction of relative characteristic point, and Y is the vertical direction of the principal direction of relative characteristic point,
dXFor the response of the Haar small echo of horizontal direction, dYFor in the response of the Haar small echo of vertical direction, | | indicate absolute
Value.
It is illustrated by taking the characteristic point 1 in multiple characteristic points as an example, the corresponding scale in above-mentioned scale space of characteristic point 1
For σ1, can be with a length of 20 σ in the principal direction top of selected characteristic point 11Square area, the rectangle region as above-mentioned default size
Domain.4*4=16 sub-regions are divided into the square area, each subregion is that side length is 20 σ1/ 4=5 σ1, then each
It may include 5*5=25 pixel in subregion, that is, above-mentioned preset quantity is 25.For including in each subregion
25 pixels calculate separately each pixel and determine the pixel in the horizontal direction and vertical direction of 1 principal direction of characteristic point
The response of the Haar small echo of point, and then generate the feature descriptor of characteristic point 1.
Ruler can also be used in addition to this above by the characteristic point in target time-frequency image is extracted using SURF algorithm
It spends invariant features and converts (Scale-invariant feature transform, SIFT) algorithm, towards quick and rotation brief introduction
The default net of the preset algorithms such as (oriented FAST and rotated Brief, ORB) algorithm and convolutional neural networks
Network model extracts the characteristic point in target time-frequency image.In embodiments of the present invention, the algorithm extracted to features described above point is not made
It is specific to limit.
In an optional embodiment, above-mentioned steps S1032 carries out multiple characteristic points using K-means algorithm
Clustering processing obtains K class, can specifically include following steps.
Step S1032A chooses K characteristic point, as target feature point from multiple characteristic points.
In this step, according to certain rules, such as rule can be randomly selected or choose at equal intervals, from what is extracted
In multiple characteristic points, K characteristic point is chosen, as target feature point.
In embodiments of the present invention, the class of the corresponding feature of each target feature point, that is, K target feature point category
In K class.
Step S1032B, for other each characteristic points, determine between other characteristic points and each target centroid away from
From, and other characteristic points are added in the class where nearest target feature point.
In this step, for other each characteristic points in addition to above-mentioned target feature point, other characteristic points are calculated
The distance between target feature point each in K target feature point.Other characteristic points are added apart from other characteristic points
In the class where nearest target feature point.
In embodiments of the present invention, that the distance between other above-mentioned characteristic points and target feature point indicate is other spies
Similarity between corresponding with the target feature point feature of sign point, apart from smaller, similarity is bigger;Distance is bigger, and similarity is got over
It is small.In addition, above-mentioned distance can indicate for Euclidean distance, Minkowski Distance and manhatton distance etc., here, to above-mentioned
Distance is not especially limited.
Step S1032C determines the clustering target of K class according to cluster result.Wherein, clustering target is a better than K is measured
The Clustering Effect of class.
In an optional embodiment, using following formula, the square error and (Sum of Squared of K class are calculated
Error, SSE), the clustering target as K class.Wherein, the clustering target of K class can indicate are as follows:
SSE is the clustering target of K class, that is, above-mentioned K class square error and, k be the quantity of class, i i-th
It is a, j-th of j, AiFor i-th of class, ajFor AiIn j-th of characteristic point, dist (aj,Ai) it is characterized point ajTo AiMiddle target signature
The distance of point.
Step S1032D, judges whether clustering target is less than pre-set level threshold value.If it is not, thening follow the steps S1032E.If
It is to then follow the steps S1032F.
In this step, determining clustering target is compared with pre-set level threshold value, determines whether clustering target is small
In pre-set level threshold value.If clustering target is less than pre-set level threshold value, it is determined that complete the clustering processing to multiple characteristic points.If
Clustering target is not less than pre-set level threshold value, it is determined that complete the clustering processing to multiple characteristic points.
Step S1032E redefines such target feature point for every one kind in K class, and returns and execute step
Rapid S1032B.
In this step, when above-mentioned clustering target is not less than pre-set level threshold value, that is, clustering target is greater than or waits
When pre-set level threshold value, electronic equipment can redefine such target feature point for each of K class class,
And return and execute above-mentioned for other each characteristic points, determine the distance between other characteristic points and each target centroid, and
Other characteristic points are added to the step in the class where nearest target feature point.
Step S1032F terminates clustering processing process, obtains the target signature data of target time-frequency image.
In this step, when above-mentioned clustering target is less than pre-set level threshold value, electronic equipment can determine clustering processing
It has been completed that, terminate clustering processing process, K class after determining cluster is the target signature data of target time-frequency image.
When the above-mentioned progress clustering processing to the multiple characteristic points extracted, using K-means algorithm.In addition to this,
Other clustering algorithms can also be used, the noise application space such as based on density clusters (Density-Based Spatial
Clustering of Applications with Noise, DBSCAN) algorithm etc., in embodiments of the present invention, to above-mentioned poly-
Clustering algorithm used by class processing is not especially limited.
By carrying out clustering processing to the characteristic point in the target time-frequency image extracted, the mesh of target time-frequency image is obtained
Mark characteristic so that target signature data classification understand, accurately, it is more representative, improve the accurate of target signature data
Property, to improve the accuracy that signal identification model identifies the classification of signal to be identified, improve signal identification efficiency.
In conclusion the method provided through the embodiment of the present invention, according to the spy in the corresponding time-frequency image of unlike signal
The difference for levying data determines the type of signal using trained signal identification model, and realization is identified not using an equipment
The signal of same type, it is no longer necessary to individually be equipped with corresponding module for the signal of each type, reduce the cost of equipment, improve
Signal identification efficiency.
In an optional embodiment, for above-mentioned signal identification model, the embodiment of the invention also provides a kind of signals
The training method of identification model.Specifically as shown in figure 4, Fig. 4 is the training side of signal identification model provided in an embodiment of the present invention
A kind of flow diagram of method.This approach includes the following steps.
Step S401 obtains default training set.
In this step, when multiple sample signals correspond to sample in the above-mentioned default training set of acquisition, that is, default frequency range
The sample characteristics data of frequency image and the sample type of each sample signal.
Specifically, obtaining the sample type of multiple sample signals and each sample signal in default frequency range.Determination obtains
The corresponding sample time-frequency image of the multiple sample signals got, and extract the sample characteristics data in sample time-frequency image.
It, can be in default frequency range when obtaining the sample type of above-mentioned multiple sample signals and each sample signal
The sample signal for sending each sample type respectively receives the sample signal.For example, sample signal is above-mentioned Bluetooth signal, Wi-
Fi signal and ZigBee signal can successively send and receive Bluetooth signal, Wi-Fi signal and ZigBee letter in default frequency range
Number, using the signal received as sample signal, and mark the sample type of each sample signal.In addition, about sample signal
Sample time-frequency image is determining and the extracting method of sample characteristics data, be referred to above-mentioned target time-frequency image determination and
The extracting method of target signature data, does not specifically illustrate herein.
For may include the sample time-frequency image of different classes of sample signal in above-mentioned default training set.Except this with
Outside, presetting can also be including the sample time-frequency image of the sample signal of multiple same types in training set.It is said by taking Fig. 5 as an example
It is bright.Fig. 5-a, Fig. 5-b and Fig. 5-c are that sample signal provided in an embodiment of the present invention is the corresponding sample time-frequency figure of Bluetooth signal
Picture.Although Fig. 5-a, Fig. 5-b and Fig. 5-c are the corresponding sample time-frequency images of Bluetooth signal, exist between three obvious
Difference, if Fig. 5-a is compared with Fig. 5-b, distributing position of the amplitude of the frequency component of Bluetooth signal in sample time-frequency image
Significantly different, Fig. 5-a and Fig. 5-b are compared with Fig. 5-c, the frequency resolution and temporal resolution of the sample time-frequency image of Fig. 5-c
The frequency resolution and temporal resolution of sample time-frequency image corresponding with Fig. 5-a and Fig. 5-b are significantly different.
In embodiments of the present invention, the data in above-mentioned default training set are not especially limited.
Multiple sample characteristics data are inputted preset machine learning model by step S402 respectively, determine each sample letter
Number type.
In this step, multiple sample characteristics data in above-mentioned default training set are inputted into preset machine learning respectively
Model obtains the corresponding type of each sample signal.
In embodiments of the present invention, above-mentioned machine learning model can be neural network model or classifier etc., such as convolution
Neural network, support vector machine (Support Vector Machine, SVM).Here, not making to have to above-mentioned machine learning model
Body limits.
Step S403 calculates damage according to the sample type of the type of determining each sample signal and each sample signal
Mistake value.
In this step, according to the type of each sample signal of above-mentioned machine learning model output, and default training
The sample type of each sample signal is concentrated, penalty values are calculated.
In one embodiment, for each sample signal, the corresponding machine learning model of the sample signal can be exported
Type be compared with the sample type of the sample signal, determine machine learning model output type it is whether correct.Statistics
Error rate of the machine learning model to the type identification of sample signal.
It is also possible to other numerical value about above-mentioned penalty values.For example, penalty values can believe sample for machine learning model
Number type identification accuracy inverse.For another example penalty values can also be expressed as above-mentioned SSE, if machine learning model
The classification of output and the sample class of sample signal are identical, then are denoted as 0;If the classification and sample signal of machine learning model output
Sample class it is not identical, then be denoted as 1.Using the corresponding SSE of multiple sample signals as penalty values.In embodiments of the present invention,
Above-mentioned penalty values are not especially limited.
Step S404, judges whether penalty values are less than preset threshold.If it is not, thening follow the steps S405.If so, executing step
Rapid S406.
In this step, according to above-mentioned penalty values, determine whether machine learning model restrains.Specifically, by above-mentioned calculating
Obtained penalty values are compared with preset threshold, determine whether penalty values are less than preset threshold, and then determine machine learning mould
Whether type restrains.When penalty values are less than preset threshold, electronic equipment can determine that machine learning model restrains.When penalty values not
When less than preset threshold, that is, be greater than or equal to preset threshold when, electronic equipment can determine that machine learning model is not converged.
It is illustrated by taking above-mentioned error rate as an example, if the preset threshold of error rate is 2%.When certain primary training, calculate
The penalty values arrived are 2.1%, 2.1% > 2%, then can determine that machine learning model is not converged.When another training, calculate
The penalty values arrived are 1.9%, 1.9% < 2%, then can determine that machine learning model restrains.
In embodiments of the present invention, above-mentioned preset threshold is not especially limited.
Step S405, adjusts the parameter of preset machine learning model, and returns and execute above-mentioned steps step S402.
In this step, when above-mentioned penalty values are not less than preset threshold, that is, when machine learning model is not converged, can
To adjust the parameter of above-mentioned machine learning model, and return execute above-mentioned steps S402, that is, return execute it is above-mentioned will be multiple
The step of sample characteristics data input preset machine learning model respectively, determine the type of each sample signal.
The method of the parameter of above-mentioned adjustment machine learning model includes but is not limited to gradient descent method, reversed adjusting method etc.,
In embodiments of the present invention, the method for adjustment of parameter in machine learning model is not especially limited.
Preset machine learning model is determined as signal identification model by step S406.
In this step, it when above-mentioned penalty values are less than preset threshold, that is, when machine learning model convergence, can incite somebody to action
Preset machine learning model is determined as above-mentioned signal identification model.At this point, the machine learning model is in above-mentioned steps S104
Preparatory trained signal identification model.
By the available trained signal identification model of the training to machine learning model, trained signal is utilized
Identification model can accurately identify the classification of signal to be identified, realize and identify different types of letter using an equipment
Number, it is no longer necessary to it is individually equipped with corresponding module for the signal of each type, the cost of equipment is reduced, improves signal identification
Efficiency.
In addition, it is existing radiofrequency signal is identified when, electronic equipment in addition to according to communication protocol difference use
Other than different communication module radiofrequency signal recognitions, energy measuring can also be carried out to different types of radiofrequency signal, by setting
Fixed energy threshold identifies each radiofrequency signal.But since the determination of decision threshold is more difficult, that is, energy threshold
More difficult determination, signal-to-noise ratio is affected to energy measuring testing result and different types of radiofrequency signal is in same frequency range
The reasons such as overlapping are likely to occur, so that poor using the recognition effect of the method radiofrequency signal recognition of energy measuring.With it is existing
It is compared using energy detection method radiofrequency signal recognition, the embodiment of the present application obtains signal knowledge by training machine learning model
Using trained signal identification model to during the classification of determination signal to be identified, and decision gate is not present in other model
The problem of limit determines difficulty, and signal-to-noise ratio and signal overlap are affected to signal identification result, this makes signal identification effect
Relatively preferably.
Based on same inventive concept, according to the signal recognition method that the embodiments of the present invention provide, the present invention is implemented
Example additionally provides a kind of signal recognition device.Specifically as shown in fig. 6, Fig. 6 is signal recognition device provided in an embodiment of the present invention
A kind of structural schematic diagram.The device comprises the following modules.
First obtains module 601, for obtaining the signal to be identified in default frequency range.
First determining module 602, for determining the target time-frequency image of signal to be identified.
Extraction module 603, for extracting the target signature data of target time-frequency image.
Second determining module 604, for by target signature data input in advance trained signal identification model, determine to
The type of identification signal, wherein signal identification model is the model obtained by presetting training set training, and default training set includes
The sample of the sample characteristics data of the corresponding sample time-frequency image of multiple sample signals and each sample signal in default frequency range
Type.
Optionally, above-mentioned first determining module 602 specifically can be used for obtaining using following Short Time Fourier Transform formula
To the target time-frequency image of signal to be identified:
Gf(w, u)=∫ f (t) g (t-u) e-jwtdt
Wherein, w is the angular frequency of signal to be identified, and f is frequency, and t is time t, and u is preset time length of window u, function
GfThe value of (w, u) is the amplitude of frequency component, and ∫ dt is the integration operation to t, and function f (t) is signal to be identified, function g
It (t-u) is preset window function, function e-jwtFor complex function, e is natural constant, and j is imaginary unit.
Optionally, said extracted module 603 may include:
Extracting sub-module obtains multiple features of target time-frequency image for carrying out feature extraction to target time-frequency image
Point.
Submodule is clustered, for using K-means clustering algorithm, clustering processing is carried out to multiple characteristic points, obtains K
Class.
First determines submodule, for determining the quantity of each class includes in K class characteristic point.
Second determines submodule, and the quantity of the characteristic point for including according to each class determines in target time-frequency image
Target signature data.
Optionally, said extracted submodule specifically can be used for extracting more in target time-frequency image using SURF algorithm
A feature descriptor obtains multiple characteristic points of target time-frequency image.
Optionally, it above-mentioned second determines submodule, specifically can be used for the quantity for the characteristic point for including according to each class, structure
Build the fisrt feature point distribution map of target time-frequency image;Fisrt feature point distribution map is determined as the target in target time-frequency image
Characteristic;Or the quantity for the characteristic point for according to each class including, characteristic point that each class includes is counted in target time-frequency image
The probability of middle appearance, and according to the corresponding probability of each class, construct the second feature point distribution map of target time-frequency image;By second
Characteristic point distribution map is determined as the target signature data in target time-frequency image.
Optionally, above-mentioned signal recognition device can also include:
Second obtains module, for obtaining default training set.
Third determining module determines every for multiple sample characteristics data to be inputted preset machine learning model respectively
The type of one sample signal.
Computing module, for according to the type of determining each sample signal and the sample type of each sample signal, meter
Calculate penalty values.
Judgment module, for judging whether penalty values are less than preset threshold.
Adjust module, for the judging result in judgment module be it is no when, adjust the parameter of preset machine learning model,
It returns to execute and multiple sample characteristics data is inputted into preset machine learning model respectively, determine the type of each sample signal
Step.
4th determining module, for the judgement in judgment module be result be when, preset machine learning model is true
It is set to signal identification model.
The device provided through the embodiment of the present invention, not according to the characteristic in the corresponding time-frequency image of unlike signal
Together, it using trained signal identification model, determines the type of signal, realizes and identify different types of letter using an equipment
Number, it is no longer necessary to it is individually equipped with corresponding module for the signal of each type, the cost of equipment is reduced, improves signal identification
Efficiency.
Based on same inventive concept, according to the signal recognition method that the embodiments of the present invention provide, the present invention is implemented
Example additionally provides a kind of electronic equipment, as shown in fig. 7, a kind of structure that Fig. 7 is electronic equipment provided in an embodiment of the present invention is shown
It is intended to.The electronic equipment includes processor 701, communication interface 702, memory 703 and communication bus 704, wherein processor
701, communication interface 702, memory 703 completes mutual communication by communication bus 704;
Memory 703, for storing computer program;
Processor 701 when for executing the program stored on memory 703, realizes following steps:
Obtain the signal to be identified in default frequency range;
Determine the target time-frequency image of signal to be identified;
Extract the target signature data of target time-frequency image;
Target signature data are inputted into trained signal identification model in advance, determine the type of signal to be identified, wherein
Signal identification model is the model obtained by presetting training set training, and presetting training set includes multiple sample letters in default frequency range
Number the sample characteristics data of corresponding sample time-frequency image and the sample type of each sample signal.
The electronic equipment provided through the embodiment of the present invention, according to the characteristic in the corresponding time-frequency image of unlike signal
Difference determine the type of signal using trained signal identification model, realize and identify different type using an equipment
Signal, it is no longer necessary to be individually equipped with corresponding communication module for the signal of each type, reduce the cost of equipment, improve
Signal identification efficiency.
The communication bus that above-mentioned electronic equipment is mentioned can be Peripheral Component Interconnect standard (Peripheral Component
Interconnect, PCI) bus or expanding the industrial standard structure (Extended Industry Standard
Architecture, EISA) bus etc..The communication bus can be divided into address bus, data/address bus, control bus etc..For just
It is only indicated with a thick line in expression, figure, it is not intended that an only bus or a type of bus.
Communication interface is for the communication between above-mentioned electronic equipment and other equipment.
Memory may include random access memory (Random Access Memory, RAM), also may include non-easy
The property lost memory (Non-Volatile Memory, NVM), for example, at least a magnetic disk storage.Optionally, memory may be used also
To be storage device that at least one is located remotely from aforementioned processor.
Above-mentioned processor can be general processor, including central processing unit (Central Processing Unit,
CPU), network processing unit (Network Processor, NP) etc.;It can also be digital signal processor (Digital Signal
Processing, DSP), it is specific integrated circuit (Application Specific Integrated Circuit, ASIC), existing
It is field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete
Door or transistor logic, discrete hardware components.
Based on same inventive concept, according to the signal recognition method that the embodiments of the present invention provide, the present invention is implemented
Example additionally provides a kind of computer readable storage medium, and computer program is stored in the computer readable storage medium, described
The step of any of the above-described signal recognition method is realized when computer program is executed by processor.
Based on same inventive concept, according to the signal recognition method that the embodiments of the present invention provide, the present invention is implemented
Example additionally provides a kind of computer program product comprising instruction, when run on a computer, so that computer executes
State either signal recognition methods in embodiment.
In the above-described embodiments, can come wholly or partly by software, hardware, firmware or any combination thereof real
It is existing.When implemented in software, it can entirely or partly realize in the form of a computer program product.The computer program
Product includes one or more computer instructions.When loading on computers and executing the computer program instructions, all or
It partly generates according to process or function described in the embodiment of the present invention.The computer can be general purpose computer, dedicated meter
Calculation machine, computer network or other programmable devices.The computer instruction can store in computer readable storage medium
In, or from a computer readable storage medium to the transmission of another computer readable storage medium, for example, the computer
Instruction can pass through wired (such as coaxial cable, optical fiber, number from a web-site, computer, server or data center
User's line (DSL)) or wireless (such as infrared, wireless, microwave etc.) mode to another web-site, computer, server or
Data center is transmitted.The computer readable storage medium can be any usable medium that computer can access or
It is comprising data storage devices such as one or more usable mediums integrated server, data centers.The usable medium can be with
It is magnetic medium, (for example, floppy disk, hard disk, tape), optical medium (for example, DVD) or semiconductor medium (such as solid state hard disk
Solid State Disk (SSD)) etc..
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device,
For the embodiments such as electronic equipment, computer readable storage medium and computer program product, since it is substantially similar to method
Embodiment, so being described relatively simple, the relevent part can refer to the partial explaination of embodiments of method.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the scope of the present invention.It is all
Any modification, equivalent replacement, improvement and so within the spirit and principles in the present invention, are all contained in protection scope of the present invention
It is interior.
Claims (10)
1. a kind of signal recognition method characterized by comprising
Obtain the signal to be identified in default frequency range;
Determine the target time-frequency image of the signal to be identified;
Extract the target signature data of the target time-frequency image;
The target signature data are inputted into trained signal identification model in advance, determine the type of the signal to be identified,
Wherein, the signal identification model is the model obtained by presetting training set training, and the default training set includes described pre-
If the sample characteristics data of the corresponding sample time-frequency image of multiple sample signals and the sample class of each sample signal in frequency range
Type.
2. the method according to claim 1, wherein the target time-frequency image of the determination signal to be identified
The step of, comprising:
Using following Short Time Fourier Transform formula, the target time-frequency image of the signal to be identified is obtained:
Gf(w, u)=∫ f (t) g (t-u) e-jwtdt
Wherein, w is the angular frequency of the signal to be identified, and f is frequency, and t is time t, and u is preset time length of window u, function
GfThe value of (w, u) is the amplitude of frequency component, and ∫ dt is the integration operation to t, and function f (t) is the signal to be identified, letter
Number g (t-u) is preset window function, function e-jwtFor complex function, e is natural constant, and j is imaginary unit.
3. the method according to claim 1, wherein the target signature extracted in the target time-frequency image
The step of data, comprising:
Feature extraction is carried out to the target time-frequency image, obtains multiple characteristic points of the target time-frequency image;
Using K-means clustering algorithm, clustering processing is carried out to the multiple characteristic point, obtains K class;
Determine the quantity for the characteristic point that each class includes in K class;
According to the quantity for the characteristic point that each class includes, the target signature data in the target time-frequency image are determined.
4. according to the method described in claim 3, it is characterized in that, it is described to the target time-frequency image carry out feature extraction,
The step of obtaining multiple characteristic points of the target time-frequency image, comprising:
Using robust feature SURF algorithm is accelerated, multiple feature descriptors in the target time-frequency image are extracted, are obtained described
Multiple characteristic points of target time-frequency image.
5. according to the method described in claim 3, it is characterized in that, the quantity of the characteristic point for including according to each class, really
The step of determining the target signature data in the target time-frequency image, comprising:
According to the quantity for the characteristic point that each class includes, the fisrt feature point distribution map of the target time-frequency image is constructed;By institute
State the target signature data that fisrt feature point distribution map is determined as in the target time-frequency image;Or,
According to the quantity for the characteristic point that each class includes, counts the characteristic point that each class includes and go out in the target time-frequency image
Existing probability, and according to the corresponding probability of each class, construct the second feature point distribution map of the target time-frequency image;It will be described
Second feature point distribution map is determined as the target signature data in the target time-frequency image.
6. the method according to claim 1, wherein the signal identification model is trained using following steps
It arrives, comprising:
Obtain the default training set;
Multiple sample characteristics data are inputted into preset machine learning model respectively, determine the type of each sample signal;
According to the sample type of the type of determining each sample signal and each sample signal, penalty values are calculated;
Judge whether penalty values are less than preset threshold;
If it is not, then adjusting the parameter of preset machine learning model, it is described that multiple sample characteristics data difference is defeated to return to execution
The step of entering preset machine learning model, determining the type of each sample signal;
If so, preset machine learning model is determined as signal identification model.
7. a kind of signal recognition device characterized by comprising
First obtains module, for obtaining the signal to be identified in default frequency range;
First determining module, for determining the target time-frequency image of the signal to be identified;
Extraction module, for extracting the target signature data of the target time-frequency image;
Second determining module, described in determining the preparatory trained signal identification model of target signature data input
The type of signal to be identified, wherein the signal identification model is the model obtained by presetting training set training, described default
Training set includes sample characteristics data of the corresponding sample time-frequency image of multiple sample signals and each in the default frequency range
The sample type of sample signal.
8. device according to claim 7, which is characterized in that first determining module is specifically used for using following short
When Fourier transform formula, obtain the target time-frequency image of the signal to be identified:
Gf(w, u)=∫ f (t) g (t-u) e-jwtdt
Wherein, w is the angular frequency of the signal to be identified, and f is frequency, and t is time t, and u is preset time length of window u, function
GfThe value of (w, u) is the amplitude of frequency component, and ∫ dt is the integration operation to t, and function f (t) is the signal to be identified, letter
Number g (t-u) is preset window function, function e-jwtFor complex function, e is natural constant, and j is imaginary unit.
9. device according to claim 7, which is characterized in that the extraction module, comprising:
Extracting sub-module obtains the multiple of the target time-frequency image for carrying out feature extraction to the target time-frequency image
Characteristic point;
Submodule is clustered, for using K-means clustering algorithm, clustering processing is carried out to the multiple characteristic point, obtains K
Class;
First determines submodule, for determining the quantity of each class includes in K class characteristic point;
Second determines submodule, and the quantity of the characteristic point for including according to each class determines in the target time-frequency image
Target signature data.
10. device according to claim 9, which is characterized in that the extracting sub-module, it is steady using accelerating to be specifically used for
Feature SURF algorithm extracts multiple feature descriptors in the target time-frequency image, obtains the more of the target time-frequency image
A characteristic point.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910371624.7A CN110097011A (en) | 2019-05-06 | 2019-05-06 | A kind of signal recognition method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910371624.7A CN110097011A (en) | 2019-05-06 | 2019-05-06 | A kind of signal recognition method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110097011A true CN110097011A (en) | 2019-08-06 |
Family
ID=67446963
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910371624.7A Pending CN110097011A (en) | 2019-05-06 | 2019-05-06 | A kind of signal recognition method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110097011A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111310700A (en) * | 2020-02-27 | 2020-06-19 | 电子科技大学 | Intermediate frequency sampling sequence processing method for radiation source fingerprint feature identification |
CN111414847A (en) * | 2020-03-18 | 2020-07-14 | 中电科海洋信息技术研究院有限公司 | Frogman identification method, device, equipment and medium |
CN111464468A (en) * | 2020-03-02 | 2020-07-28 | 南方科技大学 | Signal modulation mode identification method, device, equipment and readable storage medium |
CN112188459A (en) * | 2020-09-29 | 2021-01-05 | 普联技术有限公司 | Signal identification method, device and equipment |
CN113065387A (en) * | 2021-02-03 | 2021-07-02 | 中国船级社 | Wavelet denoising method and system for pump system |
CN116578874A (en) * | 2023-07-10 | 2023-08-11 | 武汉能钠智能装备技术股份有限公司四川省成都市分公司 | Satellite signal attribute appraising method and device based on network protocol |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103559504A (en) * | 2013-11-04 | 2014-02-05 | 北京京东尚科信息技术有限公司 | Image target category identification method and device |
CN106803081A (en) * | 2017-01-25 | 2017-06-06 | 东南大学 | A kind of brain electricity sorting technique based on Multi-classifers integrated |
CN107122738A (en) * | 2017-04-26 | 2017-09-01 | 成都蓝色起源科技有限公司 | Automatic Communication Signals Recognition based on deep learning model and its realize system |
CN108600135A (en) * | 2018-04-27 | 2018-09-28 | 中国科学院计算技术研究所 | A kind of recognition methods of signal modulation mode |
CN108764114A (en) * | 2018-05-23 | 2018-11-06 | 腾讯音乐娱乐科技(深圳)有限公司 | A kind of signal recognition method and its equipment, storage medium, terminal |
CN109077714A (en) * | 2018-07-05 | 2018-12-25 | 广州视源电子科技股份有限公司 | Signal recognition method, device, equipment and storage medium |
CN109712128A (en) * | 2018-12-24 | 2019-05-03 | 上海联影医疗科技有限公司 | Feature point detecting method, device, computer equipment and storage medium |
-
2019
- 2019-05-06 CN CN201910371624.7A patent/CN110097011A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103559504A (en) * | 2013-11-04 | 2014-02-05 | 北京京东尚科信息技术有限公司 | Image target category identification method and device |
CN106803081A (en) * | 2017-01-25 | 2017-06-06 | 东南大学 | A kind of brain electricity sorting technique based on Multi-classifers integrated |
CN107122738A (en) * | 2017-04-26 | 2017-09-01 | 成都蓝色起源科技有限公司 | Automatic Communication Signals Recognition based on deep learning model and its realize system |
CN108600135A (en) * | 2018-04-27 | 2018-09-28 | 中国科学院计算技术研究所 | A kind of recognition methods of signal modulation mode |
CN108764114A (en) * | 2018-05-23 | 2018-11-06 | 腾讯音乐娱乐科技(深圳)有限公司 | A kind of signal recognition method and its equipment, storage medium, terminal |
CN109077714A (en) * | 2018-07-05 | 2018-12-25 | 广州视源电子科技股份有限公司 | Signal recognition method, device, equipment and storage medium |
CN109712128A (en) * | 2018-12-24 | 2019-05-03 | 上海联影医疗科技有限公司 | Feature point detecting method, device, computer equipment and storage medium |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111310700A (en) * | 2020-02-27 | 2020-06-19 | 电子科技大学 | Intermediate frequency sampling sequence processing method for radiation source fingerprint feature identification |
CN111310700B (en) * | 2020-02-27 | 2024-02-13 | 电子科技大学 | Intermediate frequency sampling sequence processing method for radiation source fingerprint feature recognition |
CN111464468A (en) * | 2020-03-02 | 2020-07-28 | 南方科技大学 | Signal modulation mode identification method, device, equipment and readable storage medium |
CN111414847A (en) * | 2020-03-18 | 2020-07-14 | 中电科海洋信息技术研究院有限公司 | Frogman identification method, device, equipment and medium |
CN111414847B (en) * | 2020-03-18 | 2023-07-21 | 中电科海洋信息技术研究院有限公司 | Frog person identification method, device, equipment and medium |
CN112188459A (en) * | 2020-09-29 | 2021-01-05 | 普联技术有限公司 | Signal identification method, device and equipment |
CN112188459B (en) * | 2020-09-29 | 2023-10-03 | 普联技术有限公司 | Signal identification method, device and equipment |
CN113065387A (en) * | 2021-02-03 | 2021-07-02 | 中国船级社 | Wavelet denoising method and system for pump system |
CN116578874A (en) * | 2023-07-10 | 2023-08-11 | 武汉能钠智能装备技术股份有限公司四川省成都市分公司 | Satellite signal attribute appraising method and device based on network protocol |
CN116578874B (en) * | 2023-07-10 | 2023-11-21 | 武汉能钠智能装备技术股份有限公司四川省成都市分公司 | Satellite signal attribute appraising method and device based on network protocol |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110097011A (en) | A kind of signal recognition method and device | |
CN106851573B (en) | Log path loss model-based joint weighting K nearest neighbor indoor positioning method | |
CN106504233B (en) | Unmanned plane inspection image electric power widget recognition methods and system based on Faster R-CNN | |
CN106131959B (en) | A kind of dual-positioning method divided based on Wi-Fi signal space | |
CN108520023B (en) | Thunderstorm kernel identification and tracking method based on hybrid clustering algorithm | |
CN105260698B (en) | The method and apparatus that iris image is positioned | |
WO2021077841A1 (en) | Recurrent residual network-based signal modulation and recognition method and device | |
CN110927706A (en) | Convolutional neural network-based radar interference detection and identification method | |
CN112147573A (en) | Passive positioning method based on amplitude and phase information of CSI (channel State information) | |
CN109977731B (en) | Scene identification method, scene identification equipment and terminal equipment | |
CN105303179A (en) | Fingerprint identification method and fingerprint identification device | |
CN108960404B (en) | Image-based crowd counting method and device | |
CN111541511B (en) | Communication interference signal identification method based on target detection in complex electromagnetic environment | |
CN110363076A (en) | Personal information correlating method, device and terminal device | |
CN104820718A (en) | Image classification and searching method based on geographic position characteristics and overall situation vision characteristics | |
CN102945374B (en) | Method for automatically detecting civil aircraft in high-resolution remote sensing image | |
CN103916820A (en) | Wireless indoor locating method based on access point stability degree | |
CN114189809B (en) | Indoor positioning method based on convolutional neural network and high-dimensional 5G observation feature | |
CN110049549A (en) | More fusion indoor orientation methods and its system based on WiFi fingerprint | |
CN106817677A (en) | A kind of indoor objects information identifying method, apparatus and system based on multisensor | |
CN112541481A (en) | Sea detection radar target detection method based on deep learning | |
CN109446894A (en) | The multispectral image change detecting method clustered based on probabilistic segmentation and Gaussian Mixture | |
TW202109357A (en) | Method and electronic equipment for visual positioning and computer readable storage medium thereof | |
CN110366244A (en) | A kind of WiFi fingerprint indoor orientation method | |
Zhong et al. | WiFi indoor localization based on K-means |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190806 |