CN110097011A

CN110097011A - A kind of signal recognition method and device

Info

Publication number: CN110097011A
Application number: CN201910371624.7A
Authority: CN
Inventors: 邓中亮; 綦航; 胡恩文; 朱棣; 唐诗浩; 刘延旭
Original assignee: Beijing University of Posts and Telecommunications
Current assignee: Beijing University of Posts and Telecommunications
Priority date: 2019-05-06
Filing date: 2019-05-06
Publication date: 2019-08-06

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

Signal identification method and device

Technical Field

The present invention relates to the field of signal detection technologies, and in particular, to a signal identification method and apparatus.

Background

In a conventional navigation positioning process, positioning is often performed by identifying Global Navigation Satellite System (GNSS) signals. With the continuous development of the technology, more positioning methods such as full-source navigation, opportunistic signal navigation, multi-source fusion positioning and the like appear. In these positioning methods, positioning is usually performed by identifying all available radio frequency signals in the space domain. The radio frequency signals may include various non-pilot specific signals such as digital audio broadcasts, digital television broadcast signals, amplitude and frequency modulated broadcast signals, cellular base station signals, Bluetooth (Bluetooth) signals, ZigBee (ZigBee) signals, wireless network (Wi-Fi) signals, and the like. Unlike GNSS signals, these radio frequency signals are distributed in a wider frequency band, and the modulation scheme adopted by each signal is different, which brings great difficulty to the signal identification process.

At present, when the radio frequency signals are identified, the radio frequency signals corresponding to each communication module are identified through different communication modules, so that the equipment cost is high, and the signal identification efficiency is low.

Disclosure of Invention

The embodiment of the invention aims to provide a signal identification method and a signal identification device, so as to reduce the cost of equipment and improve the signal identification efficiency. The specific technical scheme is as follows:

the embodiment of the invention provides a signal identification method, which comprises the following steps:

acquiring a signal to be identified in a preset frequency band;

determining a target time-frequency image of the signal to be identified;

extracting target characteristic data of the target time-frequency image;

inputting the target characteristic data into a pre-trained signal identification model, and determining the type of the signal to be identified, wherein the signal identification model is obtained by training through a preset training set, and the preset training set comprises sample characteristic data of sample time-frequency images corresponding to a plurality of sample signals in a preset frequency band and the sample type of each sample signal.

Optionally, the step of determining the target time-frequency image of the signal to be identified includes:

and obtaining a target time-frequency image of the signal to be identified by using the following short-time Fourier transform formula:

G_f(w,u)＝∫f(t)g(t-u)e^-jwtdt

wherein w is the angular frequency of the signal to be identified, f is the frequency, t is the time t, u is the preset time window length u, and the function G_fThe value of (w, u) is the amplitude of the frequency component, [ integral ] dt is the integral of t, and the function f (t) is the function to be identifiedSignal, function g (t-u) being a predetermined window function, function e^-jwtIs a complex variable function, e is a natural constant, and j is an imaginary unit.

Optionally, the step of extracting the target feature data in the target time-frequency image includes:

extracting the characteristics of the target time-frequency image to obtain a plurality of characteristic points of the target time-frequency image;

clustering the plurality of feature points by adopting a K-means clustering (K-means) algorithm to obtain K classes;

determining the number of characteristic points included in each of the K classes;

and determining target characteristic data in the target time-frequency image according to the number of the characteristic points included in each class.

Optionally, the step of performing feature extraction on the target time-frequency image to obtain a plurality of feature points of the target time-frequency image includes:

and extracting a plurality of feature descriptors in the target time-frequency image by using a Speeded-Up Robust Features (SURF) algorithm to obtain a plurality of feature points of the target time-frequency image.

Optionally, the step of determining the target feature data in the target time-frequency image according to the number of the feature points included in each class includes:

constructing a first characteristic point distribution map of the target time-frequency image according to the number of the characteristic points included in each class; determining the first feature point distribution map as target feature data in the target time-frequency image; or,

counting the probability of the characteristic points included by each class in the target time-frequency image according to the number of the characteristic points included by each class, and constructing a second characteristic point distribution map of the target time-frequency image according to the probability corresponding to each class; and determining the second feature point distribution map as target feature data in the target time-frequency image.

Optionally, the signal recognition model is obtained by training through the following steps:

acquiring the preset training set;

respectively inputting the plurality of sample characteristic data into a preset machine learning model, and determining the type of each sample signal;

calculating a loss value according to the determined type of each sample signal and the determined sample type of each sample signal;

judging whether the loss value is smaller than a preset threshold value or not;

if not, adjusting parameters of a preset machine learning model, returning to execute the step of respectively inputting the plurality of sample characteristic data into the preset machine learning model and determining the type of each sample signal;

and if so, determining the preset machine learning model as a signal identification model.

An embodiment of the present invention further provides a signal identification apparatus, including:

the first acquisition module is used for acquiring a signal to be identified in a preset frequency band;

the first determining module is used for determining a target time-frequency image of the signal to be identified;

the extraction module is used for extracting target characteristic data of the target time-frequency image;

and the second determining module is used for inputting the target characteristic data into a pre-trained signal recognition model and determining the type of the signal to be recognized, wherein the signal recognition model is obtained by training through a preset training set, and the preset training set comprises sample characteristic data of sample time-frequency images corresponding to a plurality of sample signals in the preset frequency band and the sample type of each sample signal.

Optionally, the first determining module is specifically configured to obtain a target time-frequency image of the signal to be identified by using the following short-time fourier transform formula:

G_f(w,u)＝∫f(t)g(t-u)e^-jwtdt

wherein w is the angular frequency of the signal to be identified, f is the frequency, t is the time t, u is the preset time window length u, and the function G_fThe value of (w, u) is the amplitude of the frequency component, [ integral ] dt is the integral of t, the function f (t) is the signal to be identified, the function g (t-u) is a predetermined window function, and the function e^-jwtIs a complex variable function, e is a natural constant, and j is an imaginary unit.

Optionally, the extracting module includes:

the extraction submodule is used for extracting the characteristics of the target time-frequency image to obtain a plurality of characteristic points of the target time-frequency image;

the clustering submodule is used for clustering the plurality of feature points by adopting a K-means clustering algorithm to obtain K classes;

the first determining submodule is used for determining the number of the characteristic points included in each of the K classes;

and the second determining submodule is used for determining target characteristic data in the target time-frequency image according to the number of the characteristic points included in each class.

Optionally, the extracting sub-module is specifically configured to extract, by using a SURF algorithm, a plurality of feature descriptors in the target time-frequency image to obtain a plurality of feature points of the target time-frequency image.

Optionally, the second determining submodule is specifically configured to construct a first feature point distribution map of the target time-frequency image according to the number of feature points included in each class; determining the first feature point distribution map as target feature data in the target time-frequency image; or according to the number of the characteristic points included in each class, counting the probability of the characteristic points included in each class appearing in the target time-frequency image, and according to the probability corresponding to each class, constructing a second characteristic point distribution graph of the target time-frequency image; and determining the second feature point distribution map as target feature data in the target time-frequency image.

Optionally, the apparatus further comprises:

the second acquisition module is used for acquiring the preset training set;

the third determining module is used for respectively inputting the plurality of sample characteristic data into a preset machine learning model and determining the type of each sample signal;

the calculation module is used for calculating a loss value according to the determined type of each sample signal and the determined sample type of each sample signal;

the judging module is used for judging whether the loss value is smaller than a preset threshold value or not;

the adjusting module is used for adjusting the parameters of a preset machine learning model when the judging result of the judging module is negative, returning to execute the step of respectively inputting the plurality of sample characteristic data into the preset machine learning model and determining the type of each sample signal;

and the fourth determining module is used for determining a preset machine learning model as a signal recognition model when the judgment result of the judging module is yes.

The embodiment of the invention also provides electronic equipment which comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus;

a memory for storing a computer program;

and a processor for implementing any of the above signal identification method steps when executing the program stored in the memory.

An embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method implements any of the above signal identification method steps.

Embodiments of the present invention also provide a computer program product containing instructions, which when run on a computer, cause the computer to perform any of the above-mentioned signal identification methods.

The signal identification method and device provided by the embodiment of the invention can acquire a signal to be identified in a preset frequency band, determine a target time-frequency image of the signal to be identified, extract target characteristic data of the target time-frequency image, input the target characteristic data into a pre-trained signal identification model, and determine the type of the signal to be identified, wherein the signal identification model is obtained by training through a preset training set, and the preset training set comprises sample characteristic data of sample time-frequency images corresponding to a plurality of sample signals in the preset frequency band and the sample type of each sample signal. According to the technical scheme provided by the embodiment of the invention, the type of the signal is determined by utilizing the trained signal identification model according to the difference of the characteristic data in the time-frequency images corresponding to different signals, so that different types of signals can be identified by utilizing one device, and a corresponding module does not need to be independently arranged for each type of signal, thereby reducing the cost of the device and improving the signal identification efficiency.

Of course, not all of the advantages described above need to be achieved at the same time in the practice of any one product or method of the invention.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a schematic flow chart of a signal identification method according to an embodiment of the present invention;

FIG. 2-a is a time-frequency image after performing short-time Fourier transform on a Wi-Fi signal according to an embodiment of the present invention;

fig. 2-b is a time-frequency image after performing short-time fourier transform on a bluetooth signal according to an embodiment of the present invention;

fig. 2-c is a time-frequency image obtained by performing short-time fourier transform on a ZigBee signal according to an embodiment of the present invention;

FIG. 3-a is a schematic diagram of a scale space of a target time-frequency image according to an embodiment of the present invention;

FIG. 3-b is a schematic diagram of the principal direction determination of feature points provided by the embodiment of the present invention;

FIG. 4 is a schematic flow chart of a training method of a signal recognition model according to an embodiment of the present invention;

fig. 5-a is a diagram illustrating a sample time-frequency image corresponding to a bluetooth signal according to an embodiment of the present invention;

FIG. 5-b is a diagram illustrating a sample time-frequency image corresponding to a Bluetooth signal according to an embodiment of the present invention;

FIG. 5-c is a diagram illustrating a sample time-frequency image corresponding to a Bluetooth signal according to an embodiment of the present invention;

fig. 6 is a schematic structural diagram of a signal identification apparatus according to an embodiment of the present invention;

fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

At present, when radio frequency signals are identified, different communication modules are used to identify the radio frequency signals of corresponding types according to different communication protocols adopted by the radio frequency signals of different types. Because different communication models are independent of each other, the integration and the depth fusion of the equipment are not facilitated, and the cost of the equipment for identifying the signal type is higher. In addition, each communication module can only identify the signal of the corresponding type of the communication module, resulting in low efficiency of signal identification.

In order to solve the problems of high equipment cost and low signal identification efficiency, the embodiment of the invention provides a signal identification method. The method is applied to any electronic equipment comprising a navigation system or a positioning system. In the method provided by the embodiment of the invention, a signal to be recognized in a preset frequency band can be obtained, a target time-frequency image of the signal to be recognized is determined, target characteristic data of the target time-frequency image is extracted, the target characteristic data is input into a pre-trained signal recognition model, and the type of the signal to be recognized is determined, wherein the signal recognition model is obtained by training through a preset training set, and the preset training set comprises sample characteristic data of sample time-frequency images corresponding to a plurality of sample signals in the preset frequency band and the sample type of each sample signal.

According to the technical scheme provided by the embodiment of the invention, the type of the signal is determined by utilizing the trained signal identification model according to the difference of the characteristic data in the time-frequency images corresponding to different signals, so that different types of signals can be identified by utilizing one device, and a corresponding module does not need to be independently arranged for each type of signal, thereby reducing the cost of the device and improving the signal identification efficiency.

The following examples illustrate the present invention.

As shown in fig. 1, fig. 1 is a schematic flow chart of a signal identification method according to an embodiment of the present invention. The method comprises the following steps.

Step S101, obtaining a signal to be identified in a preset frequency band.

In this step, the electronic device obtains a radio frequency signal within a preset frequency band as a signal to be identified.

The radio frequency signal may include a bluetooth signal, a Wi-Fi signal, a ZigBee signal, and the like, and the preset frequency band may be a 2.4 gigahertz (GHz) Industrial Scientific Medical (ISM) frequency band, or a 5.0GHz frequency band. For example, the predetermined frequency band may be a 2.4GHz frequency band. At a certain moment, the electronic device can acquire a radio frequency signal in a 2.4GHz frequency band as a signal to be identified.

The preset frequency band can be set according to the actual application scene, the requirements of the user and the like. For example, when the broadcast signal is used for navigation or positioning, the broadcast signal may include a digital audio broadcast signal, a digital television broadcast signal, an fm broadcast signal, an am broadcast signal, and the like, and the broadcast signals are distributed in different frequency bands, so that the preset frequency band may be a frequency band corresponding to each broadcast signal. In the embodiment of the present invention, the preset frequency bands and the number of the preset frequency bands are not particularly limited.

And step S102, determining a target time-frequency image of the signal to be identified.

In this step, the signal to be recognized obtained in step S101 is a time domain signal, and the electronic device may perform a certain transformation on the signal to be recognized, that is, convert the time domain signal into a frequency domain signal, so as to obtain a time-frequency image of the signal to be recognized, which is used as a target recognition image.

In an optional embodiment, the step of determining the target time-frequency image of the signal to be recognized may be represented as:

and obtaining a target time-frequency image of the signal to be identified by using the following short-time Fourier transform formula, namely performing short-time Fourier transform on the signal to be identified by using the following formula to obtain the target time-frequency image of the signal to be identified. Wherein, the short-time fourier formula can be expressed as:

G_f(w,u)＝∫f(t)g(t-u)e^-jwtdt

w is the angular frequency of the signal to be identified, f is the frequency, t is the time t, u is the preset time window length u, the function G_fThe value of (w, u) is the amplitude of the frequency component, [ integral ] dt is the integral of t, the function f (t) is the signal to be identified, the function g (t-u) is a predetermined window function, and the function e^-jwtIs a complex variable function, e is a natural constant, and j is an imaginary unit.

In the embodiment of the invention, because the signal to be identified is often a non-stationary random signal, if the signal to be identified is directly subjected to fourier transform, a result after the fourier transform has a certain error. In order to improve the accuracy of the target time-frequency image, a preset window function is utilized to perform short-time Fourier transform on a signal to be identified. That is, when the signal to be identified is converted from the time domain signal to the frequency domain signal, the signal in the time window is intercepted from the signal to be identified through a smaller time window, and the intercepted signal is subjected to fourier transform. Due to the fact that the length of the preset time window is small, the signal intercepted in the time window can be regarded as a stable random signal, the time window is moved on the signal to be recognized, the short-time Fourier transform process of the signal to be recognized is completed, and then the target time-frequency image can be obtained.

Wi-Fi signals, Bluetooth signals, and ZigBee signals are taken as examples. After performing short-time fourier transform on the Wi-Fi signal, the bluetooth signal, and the ZigBee signal, a time-frequency image as shown in fig. 2 can be obtained. Fig. 2-a is a time-frequency image obtained by performing short-time fourier transform on a Wi-Fi signal according to an embodiment of the present invention. Fig. 2-b is a time-frequency image obtained by performing short-time fourier transform on a bluetooth signal according to an embodiment of the present invention. Fig. 2-c are time-frequency images obtained by performing short-time fourier transform on ZigBee signals according to an embodiment of the present invention. In fig. 2-a, 2-b and 2-c, the horizontal direction represents the frequency of a signal, the vertical direction represents time, and a black area in a time-frequency image is formed by a plurality of sampling points, each of which is represented by the amplitude of a frequency component at a corresponding sampling point.

In an optional embodiment, the preset window function may select different window functions according to actual conditions, user requirements, and the like, such as a Hanning window (Hanning), a Hamming window (Hamming), a Blackman window (Blackman), and the like.

In an optional embodiment, the length of the preset time window of the preset window function may be determined according to a time resolution and a frequency resolution in the time-frequency diagram. For example, when the length of the preset time window is longer, the signal to be identified intercepted in the preset time window is longer, and the frequency resolution after short-time fourier transform is higher and the time resolution is lower. When the length of the preset time window is shorter, the intercepted signal to be identified in the preset time window is shorter, and the frequency resolution after short-time Fourier transform is lower, and the time resolution is higher.

In the embodiment of the present invention, the preset window function and the preset time window are not specifically limited.

In an optional embodiment, before determining the target time-frequency image of the signal to be recognized, that is, before performing short-time fourier transform on the signal to be recognized, the electronic device may perform certain preprocessing on the signal to be recognized. For example, the signal to be identified is filtered, and noise signals in the signal to be identified are filtered. In the embodiment of the present invention, the pretreatment process and the manner are not particularly limited.

And step S103, extracting target characteristic data of the target time-frequency image.

In this step, the electronic device may perform feature extraction on the target time-frequency image of the signal to be identified to obtain target feature data.

In an optional embodiment, a preset algorithm or a preset network model may be used to perform feature extraction on the signal to be recognized, so as to obtain target feature data of a target time-frequency image. Here, the preset algorithm and the preset network model are not specifically described.

And step S104, inputting the target characteristic data into a pre-trained signal recognition model, and determining the type of the signal to be recognized.

In the step, the extracted target characteristic data of the target time-frequency image is input into a pre-trained signal identification model. According to the target characteristic data and the structure and parameters of the signal identification model, the type of the signal to be identified can be obtained. The signal recognition model is a model obtained through training of a preset training set. The preset training set may include sample feature data of a sample time-frequency image corresponding to a plurality of sample signals within a preset frequency band, and a sample type of each sample signal.

In an embodiment, when determining the type of the signal to be recognized, the signal recognition model may determine the size of the coincidence between the target feature data and the sample feature data of different sample signals according to the extracted target feature data of the signal to be recognized and the sample feature data of the sample signals in the preset training set. And according to the determined conformity, determining the sample type of the sample signal corresponding to the sample characteristic data with the maximum conformity with the target characteristic data as the type of the signal to be identified. The conformity may be expressed by mahalanobis distance, euclidean distance, or the like, and taking euclidean distance as an example, the conformity is larger as the euclidean distance is smaller. The greater the euclidean distance, the smaller the above-described conformity.

The Wi-Fi signal, the bluetooth signal, and the ZigBee signal are still used as examples for explanation. If the signal identification model is an identification model aiming at the three signals, the sample signals are Wi-Fi signals, Bluetooth signals and ZigBee signals, and the sample characteristic data are characteristic data in sample time-frequency images corresponding to the Wi-Fi signals, the Bluetooth signals and the ZigBee signals.

In an optional embodiment, for the step S103, extracting the target feature data of the target time-frequency image may specifically include the following steps.

And step S1031, performing feature extraction on the target time-frequency image to obtain a plurality of feature points of the target time-frequency image.

In this step, the electronic device may extract a plurality of feature points in the target time-frequency image of the signal to be identified by using the preset algorithm or the preset network model, so as to obtain a plurality of feature points of the target time-frequency image.

And S1032, clustering the plurality of feature points by adopting a K-means algorithm to obtain K classes.

In this step, the electronic device may perform clustering processing on the extracted multiple feature points of the target time-frequency image by using a K-means algorithm to obtain K classes.

In step S1033, the number of feature points included in each of the K classes is determined.

In this step, for each of the K classes, the number of feature points included in the class may be counted, so as to determine the number of feature points included in each of the K classes.

Step S1034, determining target characteristic data in the target time-frequency image according to the number of the characteristic points included in each class.

In an optional embodiment, a first characteristic point distribution graph of the target time-frequency image is constructed according to the number of characteristic points included in each class; and determining the first characteristic point distribution map as target characteristic data in the target time-frequency image.

Specifically, the electronic device may construct a distribution diagram of the feature points according to the number of the feature points included in each class and the position of each feature point in the target time-frequency image, and use the distribution diagram as a first feature point distribution diagram of the target time-frequency image. According to the first feature point distribution map, the electronic device may determine target feature data of the target time-frequency image, such as a distribution of feature points in the target time-frequency image, a distribution of each class, and a number of feature points included in each class.

In another optional embodiment, according to the number of the feature points included in each class, the probability of the feature points included in each class appearing in the target time-frequency image is counted, and according to the probability corresponding to each class, a second feature point distribution map of the target time-frequency image is constructed; and determining the second characteristic point distribution map as target characteristic data in the target time-frequency image.

Specifically, the electronic device may count the probability of the feature points included in each class appearing in the target time-frequency image according to the number of the feature points included in each class, and construct a feature point distribution map according to the probability corresponding to each class, as a second feature point distribution map of the target time-frequency image. The electronic device may determine target feature data of the target time-frequency image, such as a distribution of feature points in the target time-frequency image, a distribution of each class, a number of feature points included in each class, a probability of occurrence of feature points included in each class, and the like.

The target characteristic data is obtained by extracting the characteristics of the target time-frequency image and clustering the extracted characteristic points, so that the accuracy of the target characteristic data is improved, the accuracy of the signal identification model for determining the category of the signal to be identified is improved, and the signal identification efficiency is improved.

In an optional embodiment, in the step S1031, in the process of extracting the features of the target time-frequency image to obtain the multiple feature points of the target time-frequency image, the electronic device may extract the multiple feature descriptors in the target time-frequency image by using a SURF algorithm to obtain the multiple feature points of the target time-frequency image. The method specifically comprises the following steps:

and step S1031A, determining an integral image corresponding to the target time-frequency image.

In this step, the electronic device may perform integration processing on the target time-frequency image to obtain an integral image corresponding to the target time-frequency image.

In the integral image, if the pixel point at the upper left corner of the integral image is taken as the origin of coordinates, the horizontal rightward direction is the X-axis direction, and the vertical downward direction is the Y-axis direction, the integral pixel value of the pixel point at the coordinate value (X, Y) is the sum of the pixel values of all the pixel points in a rectangular region from the pixel point to the upper left corner of the integral image, that is, the pixel value and the origin of coordinates form the sum of the pixel values of all the pixel points in the rectangular region. Specifically, it can be expressed as:

wherein, I (x, y) is the integral pixel value of the pixel point (x, y), I (x)_i,y_j) Is a pixel point (x)_i,y_j) The pixel value of (2).

And step S1031B, performing convolution processing on the integral image to obtain a scale space corresponding to the target time-frequency image. Wherein, the scale space is the representation form of the image under different resolutions.

In an optional embodiment, the electronic device may process the integral image by using a box filter, that is, continuously change the size of the box filter, and perform convolution on the changed box filter and the integral image to obtain a gaussian pyramid scale space of the target time-frequency image, which is used as the scale space of the target time-frequency image.

When the gaussian pyramid scale space is constructed, that is, when the scale space of the target time-frequency image is constructed, the SURF algorithm adopts a black-plug (Hessian) matrix determinant approximate value image, and then a Hessian matrix H (I (x, y)) of an integral pixel value of a pixel point (x, y) can be expressed as:

wherein,andis the result of the second order partial derivative of the integrated pixel value I at pixel point (x, y), I being the integrated pixel value at pixel point (x, y).

The SURF algorithm performs convolution processing on the Hessian matrix determinant approximate image by using a second-order standard Gaussian function g (x, y, sigma), and constructs a Gaussian pyramid scale space, namely the scale space of a target time-frequency image. At this time, for the pixel point (x, y), when the scale is σ, the Hessian matrix may be expressed as:

wherein L is_xx(x,y,σ)、L_xy(x,y,σ)、L_yy(x, y, σ) is the result of the convolution of the second-order partial derivatives of the second-order standard function g (x, y, σ) at pixel point (x, y) with the scale σ.

Step S1031C, determining the positions of the plurality of feature points in the scale space, and obtaining a plurality of feature points.

In this step, the position of the pixel point corresponding to the local maximum of H (x, y, σ) is selected in the scale space as the position of the feature point, and a plurality of feature points are obtained.

Taking fig. 3-a as an example for explanation, fig. 3-a is a diagram of a target time-frequency image provided by the embodiment of the inventionA schematic of a scale space. Wherein the image 301 has a scale σ_iThe pixel point 302 is a pixel point in the image 301 of the image after the time-target time-frequency image and the second-order standard gaussian function are convolved. Above and below the image 301, there is a scale σ_i+1And σ_i-1The corresponding target time-frequency image is convolved with a second-order standard gaussian function (not shown in fig. 3-a). When determining the feature point in the image 302, for example, when determining whether the pixel 302 is the feature point, it is necessary to determine whether H (x, y, σ) at the pixel 302 is the largest of 26 adjacent pixels (at most 8 pixels adjacent to the pixel 302 in the image 301, and at most 18 pixels adjacent to the pixel 302 in the image adjacent to the image 301). If the value of H (x, y, σ) of the pixel 302 is the largest, the pixel 302 can be determined to be a feature point. If the value of pixel H (x, y, σ) is not the maximum, it can be determined that pixel 302 is not a feature point.

In step S1031D, the principal direction of each feature point is determined.

In this step, in order to improve the robustness of the SURF algorithm to the rotation change, that is, in order to improve the robustness of the feature points extracted after the rotation of the target time-frequency image, the main direction of each feature point may be determined.

For convenience of understanding, fig. 3-b is taken as an example for illustration, and fig. 3-b is a schematic diagram of the determination of the main direction of the feature point according to the embodiment of the present invention, in a circular region taking the feature point as a center, that is, in a region 303, a sector region 304 is rotated in the region 303 according to a preset rotation angle, for example, α, response values of Haar (Haar) wavelets of all the feature points in the sector region 304 in the horizontal direction and the vertical direction are calculated, and the sector direction in which the Haar wavelet accumulated value is the largest in the sector region 304 is determined as the main direction of the feature point, wherein the angle of the sector region is a preset angle, for example, 60 °.

Step S1031E, for each feature point, determines a feature vector corresponding to the feature point according to the principal direction of the feature point, as a feature descriptor.

In this step, for each feature point, a rectangular region of a preset size is selected according to the main direction of the feature point, and the rectangular region is divided into a plurality of sub-regions. And counting the response values of the Haar wavelets of the preset number of pixel points in each sub-region in the horizontal direction and the vertical direction. And generating a characteristic vector as a characteristic descriptor of the characteristic point according to the sum of the response values of the horizontal Haar wavelet, the sum of the absolute values of the response values of the horizontal Haar wavelet, the sum of the response values of the vertical Haar wavelet and the sum of the absolute values of the response values of the vertical Haar wavelet. Wherein the horizontal direction and the vertical direction are the horizontal direction and the vertical direction relative to the main direction of the feature point.

Feature descriptor of the above feature pointsCan be expressed as:

wherein X is a horizontal direction with respect to the principal direction of the feature point, Y is a vertical direction with respect to the principal direction of the feature point, and d_XResponse value of Haar wavelet in horizontal direction, d_YFor the response value of the Haar wavelet in the vertical direction, | · | represents an absolute value.

Taking a feature point 1 of the plurality of feature points as an example, the scale corresponding to the feature point 1 in the scale space is σ₁The length of the side in the main direction of the feature point 1 can be selected to be 20 sigma₁The square area of (2) is a rectangular area of the predetermined size. Dividing the square region into 4 × 4-16 sub-regions, each sub-region having a side length of 20 σ₁/4＝5σ₁Then, each sub-region may include 5 × 5 ═ 25 pixels, that is, the predetermined number is 25. For 25 pixel points included in each sub-regionAnd respectively calculating the response value of each pixel point in the horizontal direction and the vertical direction of the main direction of the feature point 1 to determine the Haar wavelet of the pixel point, and further generating the feature descriptor of the feature point 1.

In addition to extracting the feature points in the target time-frequency image by using the SURF algorithm, the feature points in the target time-frequency image may be extracted by using a Scale-invariant feature transform (SIFT) algorithm, a preset algorithm such as an oriented FAST and rotated Brief (ORB) algorithm, and a preset network model of a convolutional neural network. In the embodiment of the present invention, the algorithm for extracting the feature points is not particularly limited.

In an optional embodiment, for step S1032, clustering is performed on the plurality of feature points by using a K-means algorithm to obtain K classes, which may specifically include the following steps.

In step S1032A, K feature points are selected from the plurality of feature points as target feature points.

In this step, K feature points may be selected from the extracted feature points as target feature points according to a certain rule, such as random selection or equal interval selection.

In the embodiment of the present invention, each target feature point corresponds to a class of a feature, that is, K target feature points belong to K classes.

Step S1032B, for each other feature point, determining a distance between the other feature point and each target centroid, and adding the other feature point to the class in which the closest target feature point is located.

In this step, for each other feature point except the target feature point, the distance between the other feature point and each of the K target feature points is calculated. And adding the other characteristic points into the class where the target characteristic points closest to the other characteristic points are located.

In the embodiment of the present invention, the distance between the other feature point and the target feature point represents the similarity between the features corresponding to the other feature point and the target feature point, and the smaller the distance is, the greater the similarity is; the larger the distance, the smaller the similarity. The distance may be expressed by an euclidean distance, a minkowski distance, a manhattan distance, or the like, and the distance is not particularly limited.

Step S1032C, according to the clustering result, determining the clustering index of the K classes. Wherein, the clustering index is superior to the measurement of the clustering effect of K classes.

In an alternative embodiment, the Sum of Squared Errors (SSE) of the K classes is calculated as the clustering index for the K classes using the following formula. The clustering index of the K classes can be expressed as:

SSE is the clustering index of K classes, i.e. the sum of squared errors of the K classes, K is the number of classes, i is the ith, j is the jth, A_iIs the ith class, a_jIs A_iJ-th characteristic point of middle, dist (a)_j,A_i) Is a characteristic point a_jTo A_iThe distance of the middle target feature point.

Step S1032D, determine whether the clustering index is smaller than a preset index threshold. If not, step S1032E is executed. If yes, go to step S1032F.

In this step, the determined clustering index is compared with a preset index threshold value to determine whether the clustering index is smaller than the preset index threshold value. And if the clustering index is smaller than the preset index threshold, determining that the clustering processing of the plurality of feature points is finished. And if the clustering index is not less than the preset index threshold, determining to finish clustering processing of the plurality of feature points.

Step S1032E, for each of the K classes, re-determine the target feature point of the class, and return to execute step S1032B.

In this step, when the clustering index is not less than the preset index threshold, that is, when the clustering index is greater than or equal to the preset index threshold, the electronic device may re-determine the target feature point of the class for each of the K classes, and return to perform the above steps of determining the distance between the other feature point and each target centroid for each other feature point, and adding the other feature point to the class in which the target feature point closest to the other feature point is located.

Step S1032F, the clustering process is ended, and the target characteristic data of the target time-frequency image is obtained.

In this step, when the clustering index is smaller than the preset index threshold, the electronic device may determine that the clustering process is completed, end the clustering process, and determine K clustered classes as target feature data of the target time-frequency image.

When the extracted feature points are clustered, a K-means algorithm is adopted. In addition, other clustering algorithms, such as Density-Based spatial clustering of Applications with Noise (DBSCAN) algorithm, may also be used.

The extracted feature points in the target time-frequency image are clustered to obtain target feature data of the target time-frequency image, so that the target feature data are classified clearly, accurately and more typically, the accuracy of the target feature data is improved, the accuracy of the signal identification model for identifying the category of the signal to be identified is improved, and the signal identification efficiency is improved.

In summary, according to the method provided by the embodiment of the present invention, the type of the signal is determined by using the trained signal recognition model according to the difference of the feature data in the time-frequency images corresponding to different signals, so that different types of signals can be recognized by using one device, and a corresponding module does not need to be separately configured for each type of signal, thereby reducing the cost of the device and improving the signal recognition efficiency.

In an optional embodiment, for the signal recognition model, an embodiment of the present invention further provides a training method for the signal recognition model. Specifically, as shown in fig. 4, fig. 4 is a schematic flow chart of a training method of a signal recognition model according to an embodiment of the present invention. The method comprises the following steps.

Step S401, a preset training set is obtained.

In this step, the preset training set, that is, the sample characteristic data of the sample time-frequency image corresponding to the plurality of sample signals in the preset frequency band, and the sample type of each sample signal are obtained.

Specifically, a plurality of sample signals and a sample type of each sample signal are obtained within a preset frequency band. And determining sample time-frequency images corresponding to the acquired multiple sample signals, and extracting sample characteristic data in the sample time-frequency images.

When the plurality of sample signals and the sample type of each sample signal are obtained, the sample signal of each sample type may be respectively transmitted within a preset frequency band, and the sample signal may be received. For example, the sample signals are the bluetooth signal, the Wi-Fi signal and the ZigBee signal, the bluetooth signal, the Wi-Fi signal and the ZigBee signal may be sequentially transmitted and received within a preset frequency band, the received signals are used as the sample signals, and the sample type of each sample signal is marked. In addition, regarding the method for determining the sample time-frequency image of the sample signal and extracting the sample characteristic data, reference may be made to the above-mentioned method for determining the target time-frequency image and extracting the target characteristic data, which is not specifically described herein.

The preset training set may include sample time-frequency images of different classes of sample signals. In addition, the preset training set may further include a plurality of sample time-frequency images of the same type of sample signals. The description will be given by taking fig. 5 as an example. Fig. 5-a, fig. 5-b, and fig. 5-c are all sample time-frequency images corresponding to the sample signal provided by the embodiment of the present invention being a bluetooth signal. Although fig. 5-a, fig. 5-b and fig. 5-c are all sample time-frequency images corresponding to bluetooth signals, there are significant differences between the three images, for example, compared with fig. 5-a and fig. 5-b, the distribution positions of the amplitudes of the frequency components of the bluetooth signals in the sample time-frequency images are significantly different, compared with fig. 5-a and fig. 5-b, the frequency resolution and the time resolution of the sample time-frequency images of fig. 5-c are significantly different from those of the sample time-frequency images corresponding to fig. 5-a and fig. 5-b.

In the embodiment of the present invention, the data in the preset training set is not particularly limited.

Step S402, respectively inputting a plurality of sample characteristic data into a preset machine learning model, and determining the type of each sample signal.

In this step, the plurality of sample feature data in the preset training set are respectively input into a preset machine learning model, and a type corresponding to each sample signal is obtained.

In an embodiment of the present invention, the Machine learning model may be a neural network model or a classifier, such as a convolutional neural network, a Support Vector Machine (SVM). Here, the machine learning model is not particularly limited.

In step S403, a loss value is calculated according to the determined type of each sample signal and the determined sample type of each sample signal.

In this step, a loss value is calculated according to the type of each sample signal output by the machine learning model and the type of each sample signal in a preset training set.

In one embodiment, for each sample signal, the type of the machine learning model output corresponding to the sample signal may be compared with the sample type of the sample signal to determine whether the type of the machine learning model output is correct. And counting the error rate of the machine learning model on the type identification of the sample signal.

Other values for the loss value are also possible. For example, the loss value may be an inverse of a rate of correctness of the machine learning model for the type identification of the sample signal. For another example, the loss value may also be represented as the SSE, and if the class output by the machine learning model is the same as the sample class of the sample signal, it is written as 0; if the class of the output of the machine learning model is not the same as the sample class of the sample signal, it is marked as 1. And taking the SSE corresponding to the plurality of sample signals as a loss value. In the embodiment of the present invention, the loss value is not particularly limited.

Step S404, determining whether the loss value is smaller than a preset threshold. If not, go to step S405. If yes, go to step S406.

In this step, it is determined whether the machine learning model converges or not based on the above loss value. Specifically, the loss value obtained by the calculation is compared with a preset threshold value, and whether the loss value is smaller than the preset threshold value or not is determined, so that whether the machine learning model is converged or not is determined. When the loss value is less than the preset threshold, the electronic device may determine that the machine learning model converges. When the loss value is not less than the preset threshold, that is, greater than or equal to the preset threshold, the electronic device may determine that the machine learning model is not converged.

Taking the above error rate as an example, if the predetermined threshold of the error rate is 2%. When a certain training is performed, if the calculated loss value is 2.1%, and 2.1% > 2%, it can be determined that the machine learning model is not converged. On another training, the calculated loss value is 1.9%, 1.9% < 2%, and it can be determined that the machine learning model converges.

In the embodiment of the present invention, the preset threshold is not particularly limited.

Step S405, adjusting the preset parameters of the machine learning model, and returning to perform step S402.

In this step, when the loss value is not less than the preset threshold, that is, when the machine learning model does not converge, the parameters of the machine learning model may be adjusted, and the step S402 is executed again, that is, the step S is executed again to input the plurality of sample feature data into the preset machine learning model and determine the type of each sample signal.

The method for adjusting the parameters of the machine learning model includes, but is not limited to, a gradient descent method, an inverse adjustment method, and the like.

In step S406, a preset machine learning model is determined as the signal recognition model.

In this step, when the loss value is smaller than a preset threshold value, that is, when the machine learning model converges, a preset machine learning model may be determined as the signal recognition model. In this case, the machine learning model is a signal recognition model trained in advance in step S104.

The trained signal recognition model can be obtained through training of the machine learning model, the category of the signal to be recognized can be accurately recognized by the trained signal recognition model, different types of signals can be recognized by one device, corresponding modules do not need to be equipped for each type of signal independently, the cost of the device is reduced, and the signal recognition efficiency is improved.

In addition, when the existing radio frequency signals are identified, the electronic device can not only identify the radio frequency signals by adopting different communication modules according to different communication protocols, but also detect the energy of the radio frequency signals of different types, and identify each radio frequency signal through a set energy threshold. However, due to the fact that the decision threshold is difficult to determine, that is, the energy threshold is difficult to determine, the influence of the signal-to-noise ratio on the energy detection result is large, different types of radio frequency signals may overlap in the same frequency band, and the like, the identification effect of identifying the radio frequency signals by using the energy detection method is poor. Compared with the existing method for identifying the radio frequency signals by adopting the energy detection method, the method for identifying the radio frequency signals by adopting the energy detection method obtains the signal identification model by training the machine learning model, and the problems that the decision threshold is difficult to determine and the signal-to-noise ratio and the signal overlapping have large influence on the signal identification result do not exist in the process of determining the category of the signals to be identified by adopting the trained signal identification model, so that the signal identification effect is relatively good.

Based on the same inventive concept, according to the signal identification method provided by the embodiment of the invention, the embodiment of the invention also provides a signal identification device. Specifically, as shown in fig. 6, fig. 6 is a schematic structural diagram of a signal identification apparatus according to an embodiment of the present invention. The apparatus includes the following modules.

The first obtaining module 601 is configured to obtain a signal to be identified in a preset frequency band.

A first determining module 602, configured to determine a target time-frequency image of a signal to be identified.

The extracting module 603 is configured to extract target feature data of the target time-frequency image.

A second determining module 604, configured to input the target feature data into a pre-trained signal recognition model, and determine the type of the signal to be recognized, where the signal recognition model is obtained by training through a preset training set, and the preset training set includes sample feature data of sample time-frequency images corresponding to multiple sample signals in a preset frequency band, and a sample type of each sample signal.

Optionally, the first determining module 602 may be specifically configured to obtain a target time-frequency image of the signal to be identified by using the following short-time fourier transform formula:

G_f(w,u)＝∫f(t)g(t-u)e^-jwtdt

wherein w is the angular frequency of the signal to be identified, f is the frequency, t is the time t, u is the preset time window length u, and the function G_fThe value of (w, u) is the amplitude of the frequency component, and dt is the integral operation on tThe function f (t) is the signal to be identified, the function g (t-u) is a predetermined window function, and the function e^-jwtIs a complex variable function, e is a natural constant, and j is an imaginary unit.

Optionally, the extracting module 603 may include:

and the extraction submodule is used for extracting the characteristics of the target time-frequency image to obtain a plurality of characteristic points of the target time-frequency image.

And the clustering submodule is used for clustering the plurality of feature points by adopting a K-means clustering algorithm to obtain K classes.

And the first determining submodule is used for determining the number of the characteristic points included in each of the K classes.

Optionally, the extracting sub-module may be specifically configured to extract a plurality of feature descriptors in the target time-frequency image by using a SURF algorithm, so as to obtain a plurality of feature points of the target time-frequency image.

Optionally, the second determining submodule may be specifically configured to construct a first feature point distribution map of the target time-frequency image according to the number of feature points included in each class; determining the first feature point distribution map as target feature data in a target time-frequency image; or according to the number of the characteristic points included in each class, counting the probability of the characteristic points included in each class appearing in the target time-frequency image, and according to the probability corresponding to each class, constructing a second characteristic point distribution graph of the target time-frequency image; and determining the second characteristic point distribution map as target characteristic data in the target time-frequency image.

Optionally, the signal identification apparatus may further include:

and the second acquisition module is used for acquiring a preset training set.

And the third determining module is used for respectively inputting the plurality of sample characteristic data into a preset machine learning model and determining the type of each sample signal.

And the calculation module is used for calculating the loss value according to the determined type of each sample signal and the determined sample type of each sample signal.

And the judging module is used for judging whether the loss value is smaller than a preset threshold value or not.

And the adjusting module is used for adjusting the parameters of the preset machine learning model when the judging result of the judging module is negative, returning to execute the step of respectively inputting the plurality of sample characteristic data into the preset machine learning model and determining the type of each sample signal.

And the fourth determining module is used for determining the preset machine learning model as the signal recognition model when the judgment result of the judging module is yes.

According to the device provided by the embodiment of the invention, the type of the signal is determined by using the trained signal recognition model according to the difference of the characteristic data in the time-frequency images corresponding to different signals, so that different types of signals can be recognized by using one device, and a corresponding module does not need to be independently equipped for each type of signal, thereby reducing the cost of the device and improving the signal recognition efficiency.

Based on the same inventive concept, according to the signal identification method provided by the embodiment of the present invention, an embodiment of the present invention further provides an electronic device, as shown in fig. 7, and fig. 7 is a schematic structural diagram of the electronic device provided by the embodiment of the present invention. The electronic device comprises a processor 701, a communication interface 702, a memory 703 and a communication bus 704, wherein the processor 701, the communication interface 702 and the memory 703 complete mutual communication through the communication bus 704;

a memory 703 for storing a computer program;

the processor 701 is configured to implement the following steps when executing the program stored in the memory 703:

acquiring a signal to be identified in a preset frequency band;

determining a target time-frequency image of a signal to be identified;

extracting target characteristic data of a target time-frequency image;

inputting the target characteristic data into a pre-trained signal identification model, and determining the type of a signal to be identified, wherein the signal identification model is obtained by training through a preset training set, and the preset training set comprises sample characteristic data of sample time-frequency images corresponding to a plurality of sample signals in a preset frequency band and the sample type of each sample signal.

According to the electronic equipment provided by the embodiment of the invention, the type of the signal is determined by utilizing the trained signal identification model according to the difference of the characteristic data in the time-frequency images corresponding to different signals, so that different types of signals can be identified by utilizing one piece of equipment, and a corresponding communication module does not need to be independently arranged for each type of signal, thereby reducing the cost of the equipment and improving the signal identification efficiency.

The communication bus mentioned in the electronic device may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus.

The communication interface is used for communication between the electronic equipment and other equipment.

The Memory may include a Random Access Memory (RAM) or a Non-Volatile Memory (NVM), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the processor.

The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, discrete hardware component.

Based on the same inventive concept, according to the signal identification method provided by the above embodiment of the present invention, an embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, and the computer program, when executed by a processor, implements the steps of any of the signal identification methods described above.

Based on the same inventive concept, according to the signal identification method provided by the above embodiment of the present invention, an embodiment of the present invention further provides a computer program product containing instructions, which, when run on a computer, causes the computer to execute any of the signal identification methods in the above embodiments.

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, server, or data center to another website site, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.

It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

All the embodiments in the present specification are described in a related manner, and the same and similar parts among the embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for embodiments such as the apparatus, the electronic device, the computer-readable storage medium, and the computer program product, since they are substantially similar to the method embodiments, the description is simple, and for relevant points, reference may be made to part of the description of the method embodiments.

The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.

Claims

1. A signal identification method, comprising:

acquiring a signal to be identified in a preset frequency band;

determining a target time-frequency image of the signal to be identified;

extracting target characteristic data of the target time-frequency image;

2. The method according to claim 1, wherein the step of determining the target time-frequency image of the signal to be identified comprises:

G_f(w,u)＝∫f(t)g(t-u)e^-jwtdt

3. The method according to claim 1, wherein the step of extracting the target feature data in the target time-frequency image comprises:

clustering the plurality of feature points by adopting a K-means clustering algorithm to obtain K classes;

4. The method according to claim 3, wherein the step of extracting the features of the target time-frequency image to obtain a plurality of feature points of the target time-frequency image comprises:

and extracting a plurality of feature descriptors in the target time-frequency image by using an acceleration robust feature SURF algorithm to obtain a plurality of feature points of the target time-frequency image.

5. The method according to claim 3, wherein the step of determining the target feature data in the target time-frequency image according to the number of feature points included in each class comprises:

6. The method of claim 1, wherein the signal recognition model is trained by the steps comprising:

acquiring the preset training set;

judging whether the loss value is smaller than a preset threshold value or not;

7. A signal identifying apparatus, comprising:

8. The apparatus according to claim 7, wherein the first determining module is specifically configured to obtain the target time-frequency image of the signal to be identified by using the following short-time fourier transform formula:

G_f(w,u)＝∫f(t)g(t-u)e^-jwtdt

9. The apparatus of claim 7, wherein the extraction module comprises:

10. The apparatus according to claim 9, wherein the extracting sub-module is specifically configured to extract a plurality of feature descriptors in the target time-frequency image by using an accelerated robust features SURF algorithm, so as to obtain a plurality of feature points of the target time-frequency image.