CN109034088A - A kind of unmanned plane signal detection method and device - Google Patents

Abstract

The embodiment of the invention provides a kind of unmanned plane signal detection method and devices.Concrete scheme is as follows: wireless signal in available frequency range to be detected, as signal to be detected, according to predeterminated frequency interval, divide signal to be detected, obtain the wireless subsignal in multiple frequency ranges, as signal to be observed, for each signal to be observed, based on preset energy characteristic formula, extract the energy feature of the signal to be observed, based on preset cumulative measure feature formula, extract the accumulation measure feature of the signal to be observed, for each observation signal, using the energy feature of the observation signal and accumulation measure feature as input data, input SVM classifier trained in advance, utilize the categorised decision function in SVM classifier, determine in signal to be detected whether include unmanned plane signal.Using scheme provided in an embodiment of the present invention, the detection to unmanned plane signal in wireless environment may be implemented.

Description

A signal detection method and device for an unmanned aerial vehicle

technical field

The invention relates to the technical field of signal detection and identification in the field of wireless communication, in particular to a method and device for detecting signals of an unmanned aerial vehicle.

Background technique

With the continuous development of drone technology, the application of drones is becoming more and more extensive. UAV technology has brought convenience to production and life in many fields such as aerial photography, agricultural plant protection, express transportation, and disaster relief. At the same time, it is also accompanied by problems such as candid photography, delivery of illegal items, and interference with civil aviation take-off and landing signals. . Therefore, the detection of UAV signals in wireless environments has gradually become an important research direction.

Previous UAV detection methods are mostly for signal detection and signal identification of UAV signals in the mixed signal of UAV signals and noise signals, but as the wireless communication environment becomes increasingly complex, for example, UAVs can work in 2.4GHz ISM (Industrial Scientific Medical, industrial, scientific, medical) frequency band, but there are also Wi-Fi (Wireless Fidelity, wireless fidelity) signals, Bluetooth signals, signals from cordless phones, and Zigbee (Zigbee) signals in this frequency band. ) technology for wireless communication signals, etc. Therefore, the previous detection methods are no longer suitable for the detection of UAV signals.

Regarding signal detection, it mainly includes two parts: signal detection and signal identification. Signal detection refers to the process of performing specific analysis and processing on the intercepted or received signal to determine whether there is a signal in the target frequency band. Signal recognition refers to the process of performing deeper analysis and processing on the signal to further identify whether the signal that appears is the target signal.

In the process of signal detection, the received sequence is generally used to construct the test statistic first, and then the statistic is compared with the threshold or classification criterion. However, due to the phenomenon of multiple signal aliasing in the wireless environment and the complex electromagnetic environment for signal identification, there is no relatively mature identification method at present. Generally speaking, the existing signal detection and recognition algorithms mainly include the following four categories: matched filter detection, waveform detection, correlation matrix detection and energy detection. Among them, the implementation of matched filter detection and waveform detection requires sufficient prior information, but it is usually difficult to meet this condition in actual detection; although correlation matrix detection can be used for blind detection, the calculation is complicated, and the performance of the algorithm is in various Environment stability is poor. Compared with the above methods, energy detection has the advantages of simple implementation and low algorithm complexity, and is the most commonly used signal detection algorithm in actual detection.

In the prior art, according to the characteristics of the UAV signal as a trapezoidal wave in the frequency domain, based on the energy detection and identification method, the change of the signal gradient value at each frequency point of the frequency domain signal can be used to represent the energy of the signal , to identify the UAV signal. However, in the process of energy detection and identification method, because we only focus on energy features, the signal detection process is sensitive to noise signals. That is to say, when the UAV signal is in the case of low signal-to-noise ratio, due to the signal clutter in the wireless environment, the difference between the noisy signal and the pure noise signal will become smaller, and the detection of the UAV signal There will also be deviations. At the same time, when there are other trapezoidal-like non-UAV signals in the interference signal in the wireless environment, using the existing energy detection and identification method to only identify trapezoidal-like waves or signal gradients will also lead to unmanned aerial vehicle signals. Probing has gone awry.

Contents of the invention

The purpose of the embodiments of the present invention is to provide a UAV signal detection method and device, so as to realize the detection of UAV signals in a wireless environment. The specific technical scheme is as follows:

An embodiment of the present invention provides a UAV signal detection method, the method comprising:

Obtain the wireless signal in the frequency band to be detected as the signal to be detected;

Dividing the signal to be detected according to a preset frequency interval to obtain wireless sub-signals in multiple frequency bands as the signal to be observed;

For each of the signals to be observed, based on a preset energy characteristic formula, extracting an energy feature of the signal to be observed;

For each of the signals to be observed, based on a preset cumulant characteristic formula, extract the cumulant feature of the signal to be observed;

For each of the signals to be observed, the energy feature and the cumulant feature of the signal to be observed are used as input data, input into a pre-trained support vector machine SVM classifier, and the classification decision in the SVM classifier is used function to determine whether the signal to be detected contains a UAV signal, wherein the SVM classifier is trained based on the characteristic data of the sample signal, and the characteristic data of the sample signal includes each A sample support vector composed of energy features and cumulant features of each sample sub-signal, and a flag value corresponding to the sample sub-signal, the flag value indicating whether there is a UAV signal in the sample sub-signal.

Further, for each of the signals to be observed, based on a preset energy characteristic formula, the energy characteristics of the signal to be observed are extracted, including:

For each signal to be observed, based on Parseval's theorem, calculate the energy value corresponding to the signal to be observed as an energy feature, wherein, for the signal to be observed y _i (t) corresponding to the i-th frequency band, its The energy feature E _i is expressed as:

Y _i (f) represents the frequency domain signal corresponding to the signal to be observed y _i (t), f ₁ represents the lower boundary frequency corresponding to the i-th frequency band, f ₂ represents the upper boundary frequency corresponding to the i-th frequency band, Δf represents the The frequency sampling interval in the i-th frequency band, t represents time, and f represents frequency.

Further, for each of the signals to be observed, based on a preset cumulant characteristic formula, the cumulant feature of the signal to be observed is extracted, including:

For each of the signals to be observed, calculate the second-order cumulant, the fourth-order cumulant, and the sixth-order cumulant of the signal to be observed, wherein, for each of the signals to be observed, the second-order cumulant C ₂₁ represents for:

C ₂₁ ＝E(|y(k) ² |)-|Ey(k) ² |

y(k) represents the kth described signal to be observed, E( ) represents mathematical expectation, |·| represents a modulus symbol, and 1 in C ₂₁ represents the introduction of a conjugate complex number;

For each signal to be observed, the fourth-order cumulant C ₄₂ is expressed as:

C ₄₂ ＝E(|y(k) ⁴ |)-|Ey(k) ² | ² -2(E(|y(k) ² |)) ²

y(k) represents the kth described signal to be observed, E( ) represents mathematical expectation, |·| represents a modulus symbol, and 2 in C ₄₂ represents the introduction of two conjugated complex numbers;

For each signal to be observed, the sixth-order cumulant C ₆₃ is expressed as:

C ₆₃ ＝E(|y(k) ⁶ |)-9E(|y(k)| ⁴ )E(|y(k)| ² )-3E(y ^* (k) ³ y(k))E( y(k) ² )-3E(y ^* (k)y(k) ³ )E(y ^* (k) ² )-18E(y ^* (k) ² )E(y(k) ² )E(| y(k)| ² )-12(E(|y(k)| ² )) ³

y(k) represents the kth described signal to be observed, E(·) represents mathematical expectation, |·| represents a modulus symbol, * represents a conjugate symbol, and 3 in C ₆₃ represents the introduction of three conjugate complex numbers;

For each of the signals to be observed, according to the second-order cumulant, fourth-order cumulant, and sixth-order cumulant of the signal to be observed, according to the preset cumulant characteristic formula, determine the first cumulant of the signal to be observed feature and the second cumulant feature, as the cumulant feature, wherein, for each of the signals to be observed, the cumulant feature is expressed as:

γ ₁ represents the first cumulant feature, γ ₂ represents the second cumulant feature, y(k) represents the kth signal to be observed, C ₂₁ represents the second-order cumulant, and C ₄₂ represents the The fourth-order cumulant is described, C ₆₃ indicates the sixth-order cumulant, 1 in C ₂₁ indicates the introduction of a conjugate complex number, 2 in C ₄₂ indicates the introduction of two conjugate complex numbers, and 3 in C ₆₃ indicates the introduction of three Conjugate plurals.

Further, for each of the signals to be observed, the energy feature and the cumulant feature of the signal to be observed are used as input data, and input into a pre-trained support vector machine SVM classifier, using the SVM classifier The classification decision function of determining whether the signal to be detected contains a UAV signal, including:

For each of the signals to be observed, the energy feature and the cumulant feature of the signal to be observed are used as vector elements to generate a corresponding support vector, wherein the support vector corresponding to the nth signal to be observed Expressed as:

_{En represents the energy feature of the signal to be observed, γ n1 represents} the first cumulant feature of the signal to be observed, and γ _n2 represents the second cumulant feature of the observed signal;

Using the support vectors as input data into a pre-trained SVM classifier;

According to the classification decision function in the SVM classifier, determine the value corresponding to the signal to be observed as a decision value, wherein the classification decision function f(m _n ) is expressed as:

The value of the classification decision function f(m _n ) represents the decision value corresponding to the nth signal to be observed, sign( ) represents a sign function, and N represents the number of sample sub-signals, Indicates the dual value of the Lagrangian multiplier corresponding to the i-th sample support vector, <·> indicates the inner product operation, b ^* indicates the value of the bias, m _n ^T indicates the energy of the n-th signal to be observed The transposition of the support vector corresponding to the feature and the cumulant feature, x _i represents the energy feature of the i-th sample sub-signal and the sample support vector corresponding to the cumulant feature, and y _i represents the label value corresponding to the i-th sample sub-signal;

Based on the decision value of each signal to be observed, it is determined whether the signal to be detected contains a drone signal, and when at least one of the decision values is 1, it means that the signal to be detected is UAV signals are included, and when the decision values are all -1, it means that the signals to be detected do not include UAV signals.

Further, the training process of the SVM classifier includes:

Obtain a preset number of wireless signals in the frequency band to be detected according to preset conditions as sample signals;

dividing the sample signal according to a preset frequency interval to obtain wireless sub-signals in multiple frequency bands as sample sub-signals;

For each of the sample sub-signals, extracting energy features and cumulant features of the sample sub-signals;

Based on the preset number of sample signals, the sample support vector composed of the energy feature and cumulant feature of the sample sub-signal in the sample signal, and the label value corresponding to the sample sub-signal are used as input data, and the input adopts the current classification The SVM classifier with parameters and a preset structure uses the current classification decision function to classify the preset number of sample signals to complete a round of training, wherein, during the first training, the parameters of the current classifier are preset initial classifier parameters, the current classification decision function is determined according to the preset initial classifier parameters;

For each round of training, based on the preset number of sample signals, according to a preset loss function, determine the loss value of the SVM classifier;

When it is determined based on the loss value that the SVM classifier reaches a preset standard, the training is completed, and the classification decision function corresponding to the SVM classifier is determined;

When it is determined based on the loss value that the SVM classifier does not meet the preset standard, adjust the classifier parameters according to the preset adjustment method to obtain a new classification decision function, and adopt the new classification decision function and the preset Number of sample signals to complete a new round of training.

Further, for each round of training, based on the preset number of sample signals, according to a preset loss function, the loss value of the SVM classifier is determined, including:

After each round of training is completed, based on the hinge loss function, the SVM classifier loss value is determined, wherein the hinge loss function P(x) is expressed as:

The value of the hinge loss function P(x) represents the loss value, λ represents the parameter to be adjusted, w represents the normal vector of the segmentation plane in the SVM classifier, ||·|| represents the norm, and N represents the The number of sample sub-signals described above, w ^T represents the transpose of w, x _i represents the sample support vector corresponding to the energy feature and cumulant feature of the i-th sample sub-signal, y _i represents the label corresponding to the i-th sample sub-signal value, and b represents the intercept.

Further, when it is determined based on the loss value that the SVM classifier does not meet the preset standard, the parameters of the classifier are adjusted according to the preset adjustment method to obtain a new classification decision function, and the new classification decision function and the set The preset number of sample signals is used to complete a new round of training, including:

When it is determined based on the loss value that the SVM classifier does not meet the preset standard, based on the hinge loss function, using a gradient descent method to determine the value of the parameter to be adjusted in the hinge loss function;

Re-determine the classification decision function corresponding to the SVM classifier according to the value of the parameter to be adjusted;

Classify the preset number of sample signals according to the re-determined classification decision function to complete a new round of training.

Further, according to the value of the parameter to be adjusted, re-determine the classification decision function corresponding to the SVM classifier, including:

According to the value of the parameter to be adjusted, determine the penalty parameter of the SVM classifier, wherein the penalty parameter represents the weight of adjusting the function interval and classification accuracy in the optimization direction, the penalty parameter D and the parameter to be adjusted The relationship between λ is expressed as:

Based on the penalty parameter, using a convex optimization problem, determine the division method corresponding to the segmentation surface of the SVM classifier, wherein the division method of the segmentation surface represents a method for classifying the preset number of sample signals using a hyperplane , specifically expressed as:

sty _i (w ^T x _i +b)≥1-ξ _i , i=1, 2,..., N

ξ _i ≥ 0, i = 1, 2, ..., N

st represents the constraint condition, w represents the normal vector of the hyperplane, ||·|| represents the norm, D represents the penalty parameter, N represents the number of the sample sub-signals, ξ _i represents the i-th sample support vector corresponding to The slack variable of , y _i (w ^T x _i +b) represents the interval distance from the i-th sample support vector to the segmentation plane, y _i represents the label value corresponding to the i-th sample sub-signal, w ^T represents the transposition of w, x _i represents the i-th sample support vector, b represents the intercept;

Based on the Lagrangian function and the function dualization method, the solution problem of the division method is converted, and the division method after conversion is expressed as:

0≤α _i ≤D, i=1, 2,...,N

st represents the constraint condition, α _i represents the value of the Lagrangian multiplier corresponding to the i-th sample support vector, α _j represents the value of the Lagrange multiplier corresponding to the j-th sample support vector, and y _i represents the value of the i-th sample support vector The tag value corresponding to the sample sub-signal, y _j represents the tag value corresponding to the i-th sample sub-signal, The transpose of the sample support vector corresponding to the i-th sample sub-signal, x _j represents the sample support vector corresponding to the j-th sample sub-signal, <·> represents the inner product operation, D represents the penalty parameter, and N represents the number of sample sub-signals ;

Determine the value of the Lagrange multiplier based on the SMO algorithm;

Utilize the properties of the sample support vector to determine the value of the bias, wherein the value b ^* of the bias is expressed as:

N represents the number of sample sub-signals, y _n represents the label value corresponding to the n-th sample support vector, y _i represents the label value corresponding to the i-th sample support vector, Indicates the dual value of the Lagrangian multiplier corresponding to the i-th sample support vector, Indicates the transpose of the i-th sample support vector, x _n indicates the sample support vector corresponding to the n-th sample sub-signal, <·> indicates the inner product operation;

Based on the value of the Lagrangian multiplier and the value of the bias, the classification decision function is re-determined.

The embodiment of the present invention also provides a signal detection device for unmanned aerial vehicles, the device, including:

The signal acquisition module to be detected is used to obtain the wireless signal in the frequency band to be detected as the signal to be detected;

The signal to be observed acquisition module is used to divide the signal to be detected according to the preset frequency interval, and obtain wireless sub-signals in multiple frequency bands as the signal to be observed;

An energy feature extraction module, configured to extract the energy feature of the signal to be observed based on a preset energy feature formula for each signal to be observed;

A cumulant feature extraction module, configured to extract the cumulant feature of the signal to be observed based on a preset cumulant feature formula for each of the signals to be observed;

The unmanned aerial vehicle signal determination module is used for each of the signals to be observed, using the energy feature and the cumulative feature of the signal to be observed as input data, inputting a pre-trained SVM classifier, using the SVM The classification decision function in the classifier determines whether the unmanned aerial vehicle signal is included in the signal to be detected, wherein the SVM classifier is trained based on the feature data of the sample signal, and the feature data of the sample signal includes A sample support vector formed by the energy feature and cumulant feature of each sample sub-signal in the sample signal, and a flag value corresponding to the sample sub-signal, the flag value indicating whether there is no one in the sample sub-signal machine signal.

Further, the device also includes:

The sample signal acquisition module is used to acquire a preset number of wireless signals in the frequency band to be detected as sample signals according to preset conditions;

A sample sub-signal acquisition module, configured to divide the sample signal according to a preset frequency interval, and obtain wireless sub-signals in multiple frequency bands as sample sub-signals;

A feature data acquisition module, configured to extract energy features and cumulant features of the sample sub-signal for each of the sample sub-signals;

A sample signal classification module, configured to use, based on the preset number of sample signals, a sample support vector composed of energy features and cumulant features of sample sub-signals in the sample signal, and a marker value corresponding to the sample sub-signal as Input data, input an SVM classifier using current classifier parameters and a preset structure, use the current classification decision function to classify the preset number of sample signals, and complete a round of training, wherein, during the first training, the said The current classifier parameters are preset initial classifier parameters, and the current classification decision function is determined according to the preset initial classifier parameters;

A loss value determination module, configured to determine the loss value of the SVM classifier according to a preset loss function based on the preset number of sample signals for each round of training;

A classifier generating module, configured to complete training when determining that the SVM classifier reaches a preset standard based on the loss value, and determine a classification decision function corresponding to the SVM classifier;

A classifier parameter adjustment module, configured to adjust the classifier parameters according to a preset adjustment method when it is determined based on the loss value that the SVM classifier does not reach a preset standard, to obtain a new classification decision function, and adopt a new Classify the decision function and the preset number of sample signals to complete a new round of training.

A UAV signal detection method and device provided by the embodiments of the present invention can obtain wireless signals in frequency bands to be detected as signals to be detected, divide the signals to be detected according to preset frequency intervals, and obtain wireless sub-bands in multiple frequency bands. Signal, as the signal to be observed, for each signal to be observed, based on the preset energy characteristic formula, extract the energy feature of the signal to be observed, based on the preset cumulant characteristic formula, extract the cumulant feature of the signal to be observed, for each An observation signal, the energy feature and cumulant feature of the observation signal are used as input data, input into the pre-trained SVM classifier, and the classification decision function in the SVM classifier is used to determine whether the signal to be detected contains UAV signals. The detection of the UAV signal in the wireless environment can be realized through the above method.

Of course, implementing any product or method of the present invention does not necessarily need to achieve all the above-mentioned advantages at the same time.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is one of flow charts of a kind of unmanned aerial vehicle signal detection method provided by the embodiment of the present invention;

Fig. 2 is a flow chart of a SVM classifier training method for UAV signal detection provided by an embodiment of the present invention;

FIG. 3 is a schematic diagram of classifying sample signals by an SVM classifier provided by an embodiment of the present invention;

Fig. 4 is the flowchart of a kind of unmanned aerial vehicle signal detecting method provided by the embodiment of the present invention;

FIG. 5 is one of the structural schematic diagrams of a UAV signal detection device provided by an embodiment of the present invention;

Fig. 6 is the second structural schematic diagram of a UAV signal detection device provided by an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The embodiment of the present invention provides a UAV signal detection method, as shown in Figure 1, may include the following steps:

Step S101, acquiring a wireless signal in a frequency band to be detected as a signal to be detected.

In this step, use antennas, signal receivers, etc. to obtain wireless signals in the frequency band to be detected in the target environment to be detected according to preset time intervals as signals to be detected.

Specifically, according to the currently known regulations on frequency bands used by drones, the frequency bands that can be used for drones are 840.5MHz to 845MHz, 1430MHz to 1444MHz, and 2408MHz to 2440MHz. The UAV signal in the frequency band is detected, and the frequency band to be detected is 2408MHz to 2440MHz. Every 2 minutes, the signal receiver is used to collect the wireless signal in the target environment to be detected, and the collected wireless signal is used as the target environment to be detected. probe signal.

Step S102, divide the signal to be detected according to the preset frequency interval, and obtain wireless sub-signals in multiple frequency bands as the signal to be observed.

In this step, according to the spectrum diagram of the signal to be detected, the signal to be detected can be divided into equal frequency intervals to obtain wireless sub-signals corresponding to each frequency band as the signal to be observed.

Step S103, for each signal to be observed, extract the energy feature of the signal to be observed based on a preset energy feature formula.

In this step, for each signal to be observed, based on Parseval's theorem, the energy value corresponding to the signal to be observed is calculated as an energy feature.

Specifically, for the i-th signal to be observed y _i (t), using Parseval's theorem, its energy feature E _i can be calculated, where E _i can be expressed as:

Y _i (f) represents the frequency domain signal corresponding to the signal to be observed y _i (t), f ₁ represents the lower boundary frequency corresponding to the i-th frequency band, f ₂ represents the upper boundary frequency corresponding to the i-th frequency band, Δf represents the The frequency sampling interval in the i-th frequency band, t represents time, and f represents frequency.

Step S104 , for each signal to be observed, based on a preset cumulant characteristic formula, extract the cumulant feature of the signal to be observed.

In this step, for each signal to be observed, based on the second-order cumulant, fourth-order cumulant, and sixth-order cumulant corresponding to the observed signal, the first cumulant of the signal to be observed is extracted according to the preset cumulant formula feature and the second cumulant feature.

Specifically, for the kth signal to be observed y(k), its second-order cumulant C ₂₁ can be expressed as:

C ₂₁ ＝E(|y(k) ² |)-|Ey(k) ² |

Its fourth-order cumulant C ₄₂ can be expressed as:

C ₄₂ ＝E(|y(k) ⁴ |)-|Ey(k) ² | ² -2(E(|y(k) ² |)) ²

Its sixth-order cumulant C ₆₃ can be expressed as:

C ₆₃ ＝E(|y(k) ⁶ |)-9E(|y(k)| ⁴ )E(|y(k)| ² )-3E(y ^* (k) ³ y(k))E( y(k) ² )-3E(y ^* (k)y(k) ³ )E(y ^* (k) ² )-18E(y ^* (k) ² )E(y(k) ² )E(| y(k)| ² )-12(E(|y(k)| ² )) ³

In the calculation formulas of the above-mentioned second-order cumulant, fourth-order cumulant and sixth-order cumulant, y(k) represents the kth signal to be observed, E( ) represents mathematical expectation, |·| represents the modulus symbol, * Indicates the conjugate symbol, 1 in C ₂₁ means to introduce a conjugate complex number, 2 in C ₄₂ means to introduce two conjugate complex numbers, and 3 in C ₆₃ means to introduce three conjugate complex numbers.

According to the above-mentioned second-order cumulant, fourth-order cumulant, and sixth-order cumulant, the cumulant feature of each signal to be observed can be extracted according to a preset cumulant formula. Among them, the preset cumulant feature can be expressed as:

γ ₁ represents the first cumulant feature, γ ₂ represents the second cumulant feature, y(k) represents the kth signal to be observed, C ₂₁ represents the second-order cumulant, C ₄₂ represents the fourth-order cumulant, C ₆₃ represents the sixth-order cumulant The 1 in C ₂₁ means to introduce a conjugate complex number, the 2 in C ₄₂ means to introduce two conjugate complex numbers, and the 3 in C ₆₃ means to introduce three conjugate complex numbers.

It can be known from the above that since the first-order cumulant of a Gaussian random variable is the mean value of its random variable, the second-order cumulant is the variance of its random variable, and its third-order cumulant or the cumulant above the third order is equal to zero, therefore, high The order cumulant pair can have a good suppression effect on Gaussian noise. All the noise signals in the wireless environment can be suppressed by adopting the embodiments of the present invention.

In the embodiment of the present invention, for each signal to be observed, the cumulant feature of the signal to be observed mainly refers to the first cumulant feature and the first cumulant feature obtained according to its second-order cumulant, fourth-order cumulant, and sixth-order cumulant Two cumulant features, therefore, similarly, a higher-order cumulant can be introduced, and more cumulant features can be obtained according to the preset cumulant formula, for example, an eighth-order cumulant can be introduced, and the third cumulant feature can be calculated.

Step S105, for each signal to be observed, the energy feature and cumulant feature of the signal to be observed are used as input data, input into the pre-trained SVM classifier, and the classification decision function in the SVM classifier is used to determine whether the signal to be detected is Contains drone signals.

In this step, for each signal to be observed, the energy feature and cumulant feature of the signal to be observed are used as vector elements to generate corresponding support vectors, which are input into the pre-trained SVM classifier, and the classification decision function of the classifier is used Determine whether there is a UAV signal in the signal to be detected.

Specifically, for each signal to be observed, the energy feature and cumulant feature of the signal to be observed are used as vector elements to generate a corresponding support vector, where the support vector corresponding to the nth signal to be observed is Expressed as:

E _n represents the energy feature of the signal to be observed, γ _n1 represents the first cumulant feature of the signal to be observed, and γ _n2 represents the second cumulant feature of the observed signal.

The support vector is used as the input data and input into the pre-trained SVM classifier.

According to the classification decision function in the SVM classifier, determine the value corresponding to the signal to be observed as the decision value, where the classification decision function f(m _n ) is expressed as:

The value of the classification decision function f(m _n ) represents the decision value corresponding to the nth signal to be observed, sign( ) represents the sign function, N represents the number of sample sub-signals, Indicates the dual value of the Lagrangian multiplier corresponding to the support vector of the i-th sample, <·> indicates the inner product operation, b ^* indicates the value of the bias value, m _n ^T indicates the energy characteristic sum of the n-th signal to be observed The transpose of the support vector corresponding to the cumulant feature, x _i represents the energy feature of the i-th sample sub-signal and the sample support vector corresponding to the cumulant feature, and y _i represents the label value corresponding to the i-th sample sub-signal.

Based on the decision value of each signal to be observed, it is determined whether the signal to be detected contains a UAV signal. When at least one of the decision values is 1, it means that the signal to be detected contains a UAV signal. When the decision When the values are all -1, it means that the signals to be detected do not contain UAV signals.

Further, since each signal to be observed is obtained by segmenting the signal to be detected, by judging whether each signal to be observed contains a UAV signal, it can be determined whether the signal to be detected contains a UAV Signal. Therefore, the energy feature and cumulant feature of each signal to be observed are used as vector elements to generate the corresponding support vector. Use the classification decision function in the classification decision function of the pre-trained SVM classifier to classify each signal to be observed. Since the classification decision function is a symbolic function, its output value only uses 1 and -1, so when the SVM classifier is trained You can use 1 to indicate that there is a drone signal, and -1 to indicate that there is no drone signal. When there is a UAV signal in the signal to be observed, the decision value of the SVM classifier should be 1, and when there is no UAV signal in the signal to be observed, the decision value of the SVM classifier should be -1. Therefore, for all the signals to be observed included in the signal to be detected, when at least one of its decision values is 1, it means that the signal to be detected contains UAV signals; when all of its decision values are -1, then Indicates that the signals to be detected do not contain UAV signals.

In the embodiment of the present invention, the structure of the above SVM classifier may adopt an existing network structure, for example, a feed-forward network structure. By fully connecting each signal to be observed with the sample sub-signal during training, according to the parameter calculation sequence in the above classification decision function, the recognition of the UAV signal in the signal to be observed is realized, and then the UAV signal in the signal to be detected is realized. Signal recognition.

To sum up, in the embodiment of the present invention, for the signal to be observed, the characteristics of the noise signal can be suppressed by using the high-order cumulant, and the noise signal in the signal to be observed is suppressed, and then the SVM classifier is used to identify the noise signal in the signal to be detected The jamming signal and UAV signal can realize the detection of UAV signal in the wireless environment.

In one embodiment of above-mentioned a kind of unmanned aerial vehicle signal detection method, the training process of SVM classifier in the above-mentioned step S105, as shown in Figure 2, can comprise the following steps:

Step S201, according to preset conditions, acquire a preset number of wireless signals in the frequency band to be detected as sample signals.

In this step, in the case of UAV signals, a preset number of wireless signals can be obtained in the frequency band to be detected as positive sample signals; A preset number of samples of the signal, as a negative sample signal.

Specifically, in the above-mentioned target environment, the UAV can be used to simulate the presence or absence of the UAV signal, and a preset number of positive sample signals and negative sample signals can be obtained in the frequency band to be detected. For example, assuming that the frequency band to be detected is 2408MHz to 2440MHz, in the target environment, use the UAV to send 2410MHz UAV signal, and take 30 seconds as the time interval to obtain 1000 copies of wireless signals in the frequency band to be detected as positive Sample signal; Similarly, under the same conditions, no drone sends a drone signal, and 1000 wireless signals are obtained as negative sample signals.

Step S202, divide the sample signal according to preset frequency intervals to obtain wireless sub-signals in multiple frequency bands as sample sub-signals.

In this step, for each sample signal, according to the spectrum diagram of the sample signal, the sample signal to be treated is divided into equal frequency intervals to obtain wireless sub-signals corresponding to each frequency band as sample sub-signals.

Step S203, for each sample sub-signal, extract the energy feature and cumulant feature of the sample sub-signal.

In this step, the energy feature and cumulant feature of each sample sub-signal can be determined in the same manner as the above step S103 and step S104.

Specifically, for each sample sub-signal, the energy feature of the sample sub-signal is extracted based on the preset energy feature formula, where the energy feature E _i ′ of the sample sub-signal y _i ′(t) in the i-th frequency band represents for:

Y _i ′(f) represents the frequency domain signal corresponding to the sample sub-signal y _i ′(t) in the i-th frequency band, f ₁ represents the lower boundary frequency corresponding to the i-th frequency band, and f ₂ represents the upper boundary frequency corresponding to the i-th frequency band Boundary frequency, Δf represents the sampling interval of the frequency in the i-th frequency band, t represents time, and f represents frequency.

For each sample sub-signal, according to the preset cumulant feature formula, determine the first cumulant feature and the second cumulant feature of the sample sub-signal as the cumulant feature, wherein the kth sample sub-signal y'(k ), the first cumulant feature γ ₁ ′ and the second cumulant feature γ ₂ ′ are expressed as:

C ₂₁ ′ represents the second-order cumulant of the sample sub-signal, C ₄₂ ′ represents the fourth-order cumulant of the sample sub-signal, C ₆₃ ′ represents the sixth-order cumulant of the sample sub-signal, and 1 in C ₂₁ ′ represents the introduction of a For conjugate complex numbers, 2 in C ₄₂ ′ means introducing two conjugate complex numbers, and 3 in C ₆₃ ′ means introducing three conjugate complex numbers.

Step S204, based on the preset number of sample signals, the sample support vector composed of the energy feature and cumulant feature of the sample sub-signal in the sample signal, and the label value corresponding to the sample sub-signal are used as input data, and the input adopts the current classifier parameters and The SVM classifier with a preset structure uses the current classification decision function to classify a preset number of sample signals to complete a round of training.

In the step, based on the preset number of sample signals, the corresponding sample support vectors will be generated from the energy features and cumulant features of the sample sub-signals in the sample signals, and each sample support vector and the label value of its corresponding sample sub-signal , as the input data, input into the SVM classifier with the current classifier parameters of the preset structure, use the current classification decision function to classify the preset number of sample signals, and complete a round of training.

Specifically, for one round of training of the SVM classifier, the SVM classifier needs to be used to classify all sample sub-signals corresponding to the preset number of sample signals. For example, suppose there are 1000 positive sample signals and 1000 negative sample signals above, that is, the number of sample signals is 2000. After processing in step S202, each sample signal generates 10 sample sub-signals, that is, there are 20,000 sample sub-signals. The current round of training is to use the classification decision function to complete the classification of these 20,000 sample sub-signals.

In the embodiment of the present invention, when training for the first time, the current classifier parameters of the SVM classifier are preset, and the parameters in the classification decision function are determined according to the current classifier parameters.

Step S205, for each round of training, based on a preset number of sample signals, and according to a preset loss function, determine the loss value of the SVM classifier.

In this step, after each round of training is completed, the loss value of the SVM classifier can be determined based on the hinge loss function, where the hinge loss function P(x) is expressed as:

The value of the hinge loss function P(x) represents the loss value, λ represents the parameter to be adjusted, w represents the normal vector of the segmentation surface in the SVM classifier, ||·|| represents the norm, N represents the number of sample sub-signals, w ^T represents the transpose of w, x _i represents the sample support vector corresponding to the energy feature and cumulant feature of the i-th sample sub-signal, y _i represents the label value corresponding to the i-th sample sub-signal, and b represents the intercept.

Step S206, when it is determined based on the loss value that the SVM classifier meets the preset standard, the training is completed, and the classification decision function corresponding to the SVM classifier is determined.

In this step, when the loss value is less than or equal to the preset threshold, it is considered that the SVM classifier reaches the preset standard, and the training can be ended. At this time, the SVM classifier with the preset structure is the SVM classifier in the above step S105. Currently The classification decision function is the classification decision function in the above step S105.

Step S207, when it is determined based on the loss value that the SVM classifier does not meet the preset standard, adjust the parameters of the classifier according to the preset adjustment method to obtain a new classification decision function, and use the new classification decision function and a preset number of sample signals, Complete a new round of training.

In this step, when the loss value is greater than the preset threshold, it is considered that the SVM classifier has not reached the preset standard, and the gradient of the above-mentioned hinge loss function to λ can be calculated according to the gradient descent method, and the above-mentioned current classifier parameters are adjusted to obtain a new The classification decision function, and use the new classification decision function and the preset number of sample signals to complete a new round of training.

Specifically, the training process of the above SVM classifier is to classify the sample signal, as shown in Figure ₃ , x1 and x1 represent the feature data _of the signal, for example, the first cumulant feature and the second cumulant feature, "+ "Represents a positive sample signal, "-" represents a negative sample signal. In a high-dimensional space, the positive sample signal and the negative sample signal are classified by using a hyperplane.

Among them, the hyperplane can be expressed as:

w ^T x + b = 0

w represents the normal vector of the hyperplane, w ^T represents the transpose of w, and b represents the intercept.

Then, the selection of the hyperplane should choose the middle position of the positive sample signal and the negative sample signal, that is, the positive sample signal and the negative sample signal farthest from the hyperplane are equal to the hyperplane, which can be expressed as:

sty _i (w ^T x _i +b)≥1, i=1, 2,..., N

In order to facilitate the solution of the hyperplane problem, it can be transformed into:

sty _i (w ^T x _i +b)≥1, i=1, 2,..., N

When the SVM classifier classifies the sample signal, after each round of training is completed, the plane needs to be re-determined according to the values of the above-mentioned parameters to be adjusted. Therefore, the penalty parameter D and the relaxation variable ξ are introduced, where the penalty parameter represents the function in the adjustment optimization direction The interval and the weight of the classification accuracy, the slack variable represents the amount of variation that the sample signal allows to deviate from the margin of the hyperplane.

Among them, the relationship between the penalty parameter D and the slack variable ξ and the above hinge loss function can be expressed as:

The hyperplane problem can be expressed as:

sty _i (w ^T x _i +b)≥1-ξ _i , i=1, 2,..., N

ξ _i ≥ 0, i = 1, 2, ..., N

st represents the constraint condition, w represents the normal vector of the hyperplane, ||·|| represents the norm, D represents the penalty parameter, N represents the number of sample sub-signals, ξ _i represents the slack variable corresponding to the i-th sample support vector, y _i (w ^T x _i +b) represents the interval distance from the i-th sample support vector to the segmentation plane, y _i represents the label value corresponding to the i-th sample sub-signal, w ^T represents the transpose of w, _xi represents the i sample support vectors, b represents the intercept.

Further, the Lagrangian function L(w, b, ξ, α, μ) can be constructed, which can be expressed as:

Among them, α _i ≥ 0, μ _i ≥ 0, w represents the normal vector of the hyperplane, ||·|| represents the norm, D represents the penalty parameter, N represents the number of sample sub-signals, ξ _i represents the i-th sample The slack variable corresponding to the support vector, α _i and μ _i represent the values of two Lagrangian multipliers, y _i represents the label value corresponding to the i-th sample sub-signal, w ^T represents the transpose of w, _xi represents the The sample support vector corresponding to the i-th sample sub-signal, b represents the intercept, and ξ _i represents the slack variable corresponding to the i-th sample sub-signal.

Using the dual method to transform the hyperplane problem, it can be specifically expressed as:

0≤α _i ≤D, i=1, 2,...,N

st represents the constraints, α _i represents the value of the Lagrange multiplier corresponding to the i-th sample sub-signal, α _j represents the value of the Lagrange multiplier corresponding to the j-th sample sub-signal, and y _i represents the value of the i-th sample sub-signal The tag value corresponding to the sample sub-signal, y _j represents the tag value corresponding to the i-th sample sub-signal, The transpose of the sample support vector corresponding to the i-th sample sub-signal, x _j represents the sample support vector corresponding to the j-th sample sub-signal, <·> represents the inner product operation, D represents the penalty parameter, and N represents the number of sample sub-signals .

Based on the SMO algorithm, the value of the Lagrangian multiplier is determined.

Using the properties of the sample support vector, determine the value of the bias, where the value of the bias b ^* is expressed as:

N represents the number of sample sub-signals, y _n represents the label value corresponding to the n-th sample support vector, y _i represents the label value corresponding to the i-th sample support vector, Indicates the dual value of the Lagrangian multiplier corresponding to the i-th sample support vector, Indicates the transpose of the i-th sample support vector, x _n indicates the sample support vector corresponding to the n-th sample sub-signal, <·> indicates the inner product operation.

Based on the value of the Lagrangian multiplier and the value of the bias, the classification decision function is redefined.

In the embodiment of the present invention, the above property of the sample support vector means that for each sample sub-signal, the product of the decision function mapping value of the sample support vector and its label value is 1.

To sum up, since the sample signal only includes the positive sample signal and the negative sample signal, the accuracy of the classification decision function can be better determined by using the sample signal to train the SVM classifier. Therefore, the use of the SVM classifier in the embodiment of the present invention can more accurately determine whether there is a UAV signal in the signal to be detected.

By adopting the embodiment of the present invention, the detection of the UAV signal in the wireless environment can be realized. As shown in FIG. 4, in the SVM classifier training stage, by extracting the energy feature and cumulative feature of the original signal in the above-mentioned target environment, And extract the energy special diagnosis and cumulant features from the UAV signal and the original signal, use the extracted feature data to construct a training sample set, and use the training sample set to train the SVM classifier. In the identification process, according to the energy characteristics and cumulative characteristics of the signal to be detected, the trained SVM classifier can be used to determine whether there is a UAV signal in the signal to be detected.

Based on the same inventive concept, according to a method for detecting a UAV signal provided by the above-mentioned embodiment of the present invention, an embodiment of the present invention also provides a device for detecting a UAV signal, as shown in FIG. 5 , which may include:

The signal-to-be-detected acquisition module 501 is configured to acquire a wireless signal within a frequency band to be detected as a signal to be detected.

The signal-to-be-observed acquisition module 502 is configured to divide the signal to be detected according to a preset frequency interval, and obtain wireless sub-signals in multiple frequency bands as the signal to be observed.

The energy feature extraction module 503 is configured to, for each signal to be observed, extract the energy feature of the signal to be observed based on a preset energy feature formula.

The cumulant feature extraction module 504 is configured to, for each signal to be observed, extract the cumulant feature of the signal to be observed based on a preset cumulant feature formula.

UAV signal determination module 505, for each signal to be observed, the energy feature and cumulative quantity feature of the signal to be observed are used as input data, input into the pre-trained support vector machine SVM classifier, using the The classification decision function determines whether the signal to be detected contains the UAV signal.

Further, the energy feature extraction module 503 is specifically configured to, for each signal to be observed, calculate the energy value corresponding to the signal to be observed based on Parseval's theorem as the energy feature, wherein, for the signal to be observed corresponding to the i-th frequency band Signal y _i (t), its energy feature E _i is expressed as:

Y _i (f) represents the frequency domain signal corresponding to the signal to be observed y _i (t), f ₁ represents the lower boundary frequency corresponding to the i-th frequency band, f ₂ represents the upper boundary frequency corresponding to the i-th frequency band, Δf represents the The frequency sampling interval in the i-th frequency band, t represents time, and f represents frequency.

Further, the above-mentioned cumulant feature extraction module 504 includes:

The cumulant calculation sub-module is used for calculating the second-order cumulant, the fourth-order cumulant and the sixth-order cumulant of the signal to be observed for each signal to be observed, wherein, for each signal to be observed, the second-order cumulant C ₂₁ can be expressed as:

C ₂₁ ＝E(|y(k) ² |)-|Ey(k) ² |

y(k) represents the kth signal to be observed, E(·) represents the mathematical expectation, |·| represents the modulus symbol, and 1 in C ₂₁ represents the introduction of a conjugate complex number;

For each signal to be observed, the fourth-order cumulant C ₄₂ can be expressed as:

C ₄₂ ＝E(|y(k) ⁴ |)-|Ey(k) ² | ² -2(E(|y(k) ² |)) ²

y(k) represents the kth signal to be observed, E( ) represents mathematical expectation, |·| represents a modulus symbol, and 2 in C ₄₂ represents the introduction of two conjugate complex numbers;

For each signal to be observed, the cumulant C ₆₃ can be expressed as:

C ₆₃ ＝E(|y(k) ⁶ |)-9E(|y(k)| ⁴ )E(|y(k)| ² )-3E(y ^* (k) ³ y(k))E( y(k) ² )-3E(y ^* (k)y(k) ³ )E(y ^* (k) ² )-18E(y ^* (k) ² )E(y(k) ² )E(| y(k)| ² )-12(E(|y(k)| ² )) ³

y(k) represents the kth signal to be observed, E(·) represents the mathematical expectation, |·| represents the modulus symbol, * represents the conjugate symbol, and the 3 in C ₆₃ represents the introduction of three conjugate complex numbers.

The cumulant feature calculation sub-module is used to determine the signal to be observed according to the second-order cumulant, fourth-order cumulant, and sixth-order cumulant of the signal to be observed according to the preset cumulant characteristic formula for each signal to be observed The first cumulant feature and the second cumulant feature of are used as the cumulant feature, where, for each signal to be observed, the cumulant feature is expressed as:

γ ₁ represents the first cumulant feature, γ ₂ represents the second cumulant feature, y(k) represents the kth signal to be observed, C ₂₁ represents the second-order cumulant, C ₄₂ represents the fourth-order cumulant, C ₆₃ represents the sixth-order cumulant The 1 in C ₂₁ means to introduce a conjugate complex number, the 2 in C ₄₂ means to introduce two conjugate complex numbers, and the 3 in C ₆₃ means to introduce three conjugate complex numbers.

Further, the above-mentioned UAV signal determination module 505 includes:

The support vector generating submodule is used for each signal to be observed, using the energy feature and cumulant feature of the signal to be observed as vector elements to generate a corresponding support vector, wherein the support vector corresponding to the nth signal to be observed Expressed as:

E _n represents the energy feature of the signal to be observed, γ _n1 represents the first cumulant feature of the signal to be observed, and γ _n2 represents the second cumulant feature of the observed signal.

The support vector input submodule is used to input the support vector as input data into the pre-trained SVM classifier.

The decision value determination sub-module is used to determine the value corresponding to the signal to be observed according to the classification decision function in the SVM classifier as the decision value, wherein the classification decision function f(m _n ) is expressed as:

The value of the classification decision function f(m _n ) represents the decision value corresponding to the nth signal to be observed, sign( ) represents the sign function, N represents the number of sample sub-signals, Indicates the dual value of the Lagrangian multiplier corresponding to the support vector of the i-th sample, <·> indicates the inner product operation, b ^* indicates the value of the bias value, m _n ^T indicates the energy characteristic sum of the n-th signal to be observed The transpose of the support vector corresponding to the cumulant feature, x _i represents the energy feature of the i-th sample sub-signal and the sample support vector corresponding to the cumulant feature, and y _i represents the label value corresponding to the i-th sample sub-signal.

The UAV signal determination sub-module is used to determine whether the UAV signal is included in the signal to be detected based on the decision value of each signal to be observed. When at least one of the decision values has a value of 1, it means that the signal to be detected is The signal contains UAV signals, and when the decision values are all -1, it means that the signal to be detected does not contain UAV signals.

Further, the UAV signal detection device provided by the above-mentioned embodiment of the present invention, as shown in FIG. 6 , may also include:

The sample signal acquisition module 601 is configured to acquire a preset number of wireless signals in the frequency band to be detected as sample signals according to preset conditions.

The sample sub-signal acquisition module 602 is configured to divide the sample signal according to preset frequency intervals to obtain wireless sub-signals in multiple frequency bands as sample sub-signals.

The feature data acquisition module 603 is configured to, for each sample sub-signal, extract the energy feature and cumulant feature of the sample sub-signal.

The sample signal classification module 604 is configured to use a sample support vector composed of the energy feature and cumulant feature of the sample sub-signal in the sample signal and the corresponding label value of the sample sub-signal as input data based on a preset number of sample signals. The current classifier parameters and the SVM classifier with the preset structure use the current classification decision function to classify the preset number of sample signals and complete a round of training. In the first training, the current classifier parameters are the preset initial classification The current classification decision function is determined according to the preset initial classifier parameters.

The loss value determination module 605 is configured to determine the loss value of the SVM classifier according to a preset loss function based on a preset number of sample signals for each round of training.

The classifier generating module 606 is configured to complete the training and determine the classification decision function corresponding to the SVM classifier when it is determined based on the loss value that the SVM classifier meets the preset standard.

The classifier parameter adjustment module 607 is used to adjust the classifier parameters according to the preset adjustment method when it is determined based on the loss value that the SVM classifier does not meet the preset standard, to obtain a new classification decision function, and to adopt the new classification decision function and the preset Set the number of sample signals to complete a new round of training.

Further, the loss value determination module 605 is specifically configured to determine the loss value of the SVM classifier based on the hinge loss function after each round of training is completed, wherein the hinge loss function P(x) is expressed as:

The value of the hinge loss function P(x) represents the loss value, λ represents the parameter to be adjusted, w represents the normal vector of the segmentation surface in the SVM classifier, ||·|| represents the norm, N represents the number of sample sub-signals, w ^T represents the transpose of w, x _i represents the sample support vector corresponding to the energy feature and cumulant feature of the i-th sample sub-signal, y _i represents the label value corresponding to the i-th sample sub-signal, and b represents the intercept.

Further, the above-mentioned classifier parameter adjustment module 607 includes:

The sub-module for determining parameters to be adjusted is used to determine the value of the parameters to be adjusted in the hinge loss function based on the hinge loss function and using the gradient descent method when it is determined that the SVM classifier does not meet the preset standard based on the loss value.

The classification decision function determination sub-module is used to re-determine the classification decision function corresponding to the SVM classifier according to the value of the parameter to be adjusted.

Further, the above-mentioned classification decision function determination submodule also includes:

The penalty parameter determination submodule is used to determine the penalty parameter of the SVM classifier according to the value of the parameter to be adjusted, wherein the penalty parameter represents the weight of adjusting the function interval and classification accuracy in the optimization direction, and the difference between the penalty parameter D and the parameter λ to be adjusted The relationship between is expressed as:

The hyperplane determination submodule is used to determine the division method of the SVM classifier corresponding to the segmentation surface based on the penalty parameter and using the convex optimization problem, wherein the division method of the segmentation surface represents the method of classifying a preset number of sample signals using the hyperplane , specifically expressed as:

sty _i (w ^T x _i +b)≥1-ξ _i , i=1, 2,..., N

ξ _i ≥ 0, i = 1, 2, ..., N

st represents the constraint condition, w represents the normal vector of the hyperplane, ||·|| represents the norm, D represents the penalty parameter, N represents the number of sample sub-signals, ξi represents the slack variable corresponding to the i-th sample sub-signal, y _i (w ^T x _i +b) represents the interval distance from the i-th sample support vector to the segmentation plane, y _i represents the label value corresponding to the i-th sample sub-signal, w ^T represents the transpose of w, and _xi represents the i-th sample support vectors, and b represents the intercept.

The problem conversion sub-module is used to solve the problem based on the Lagrangian function and the function dualization method, and converts the partition method. The converted partition method is expressed as:

0≤α _i ≤D, i=1, 2,...,N

st represents the constraints, α _i represents the value of the Lagrange multiplier corresponding to the i-th sample sub-signal, α _j represents the value of the Lagrange multiplier corresponding to the j-th sample sub-signal, and y _i represents the value of the i-th sample sub-signal The tag value corresponding to the sample sub-signal, y _j represents the tag value corresponding to the i-th sample sub-signal, The transpose of the sample support vector corresponding to the i-th sample sub-signal, x _j represents the sample support vector corresponding to the j-th sample sub-signal, <·> represents the inner product operation, D represents the penalty parameter, and N represents the number of sample sub-signals ;

The Lagrangian multiplier determining submodule is used to determine the value of the Lagrangian multiplier based on the SMO algorithm.

The offset determination sub-module is used to determine the value of the offset by using the properties of the sample support vector, wherein the value b ^* of the offset is expressed as:

N represents the number of sample sub-signals, y _n represents the label value corresponding to the n-th sample support vector, y _i represents the label value corresponding to the i-th sample support vector, Indicates the dual value of the Lagrangian multiplier corresponding to the i-th sample support vector, Indicates the transpose of the i-th sample support vector, x _n indicates the sample support vector corresponding to the n-th sample sub-signal, <·> indicates the inner product operation;

Based on the value of the Lagrangian multiplier and the value of the bias, the classification decision function is redefined.

It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is a relationship between these entities or operations. any such actual relationship or order exists between them. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Each embodiment in this specification is described in a related manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for relevant parts, please refer to part of the description of the method embodiment.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present invention are included in the protection scope of the present invention.

Claims (10)

1. A UAV signal detection method, characterized in that, comprising: Obtain the wireless signal in the frequency band to be detected as the signal to be detected; Dividing the signal to be detected according to a preset frequency interval to obtain wireless sub-signals in multiple frequency bands as the signal to be observed; For each of the signals to be observed, based on a preset energy characteristic formula, extracting an energy feature of the signal to be observed; For each of the signals to be observed, based on a preset cumulant characteristic formula, extract the cumulant feature of the signal to be observed; For each of the signals to be observed, the energy feature and the cumulant feature of the signal to be observed are used as input data, input into a pre-trained support vector machine SVM classifier, and the classification decision in the SVM classifier is used function to determine whether the signal to be detected contains a UAV signal, wherein the SVM classifier is trained based on the characteristic data of the sample signal, and the characteristic data of the sample signal includes each A sample support vector composed of energy features and cumulant features of each sample sub-signal, and a flag value corresponding to the sample sub-signal, the flag value indicating whether there is a UAV signal in the sample sub-signal. 2. The method according to claim 1, wherein, for each of the signals to be observed, extracting the energy characteristics of the signal to be observed based on a preset energy characteristic formula includes: For each signal to be observed, based on Parseval's theorem, calculate the energy value corresponding to the signal to be observed as an energy feature, wherein, for the signal to be observed y _i (t) corresponding to the i-th frequency band, its The energy feature _Ei is expressed as: Y _i (f) represents the frequency domain signal corresponding to the signal to be observed y _i (t), f ₁ represents the lower boundary frequency corresponding to the i-th frequency band, f ₂ represents the upper boundary frequency corresponding to the i-th frequency band, Δf represents the The frequency sampling interval in the i-th frequency band, t represents time, and f represents frequency. 3. The method according to claim 1, wherein, for each of the signals to be observed, based on a preset cumulative characteristic formula, extracting the cumulative characteristic of the signal to be observed includes: For each of the signals to be observed, calculate the second-order cumulant, the fourth-order cumulant, and the sixth-order cumulant of the signal to be observed, wherein, for each of the signals to be observed, the second-order cumulant C ₂₁ represents for: C ₂₁ ＝E(|y(k) ² |)-|Ey(k) ² | y(k) represents the kth described signal to be observed, E( ) represents mathematical expectation, |·| represents a modulus symbol, and 1 in C ₂₁ represents the introduction of a conjugate complex number; For each signal to be observed, the fourth-order cumulant C ₄₂ is expressed as: C ₄₂ ＝E(|y(k) ⁴ |)-|Ey(k) ² | ² -2(E(|y(k) ² |)) ² y(k) represents the kth described signal to be observed, E( ) represents mathematical expectation, |·| represents a modulus symbol, and 2 in C ₄₂ represents the introduction of two conjugated complex numbers; For each signal to be observed, the sixth-order cumulant C ₆₃ is expressed as: C ₆₃ ＝E(|y(k) ⁶ |)-9E(|y(k)| ⁴ )E(|y(k)| ² )-3E(y ^* (k) ³ y(k))E( y(k) ² ) -3E(y ^* (k)y(k) ³ )E(y ^* (k) ² )-18E(y ^* (k) ² )E(y(k) ² )E(|y(k)| ² ) -12(E(|y(k)| ² )) ³ y(k) represents the kth described signal to be observed, E(·) represents mathematical expectation, |·| represents a modulus symbol, * represents a conjugate symbol, and 3 in C ₆₃ represents the introduction of three conjugate complex numbers; For each of the signals to be observed, according to the second-order cumulant, fourth-order cumulant, and sixth-order cumulant of the signal to be observed, according to the preset cumulant characteristic formula, determine the first cumulant of the signal to be observed feature and the second cumulant feature, as the cumulant feature, wherein, for each of the signals to be observed, the cumulant feature is expressed as: γ ₁ represents the first cumulant feature, γ ₂ represents the second cumulant feature, y(k) represents the kth signal to be observed, C ₂₁ represents the second-order cumulant, and C ₄₂ represents the The fourth-order cumulant is described, C ₆₃ indicates the sixth-order cumulant, 1 in C ₂₁ indicates the introduction of a conjugate complex number, 2 in C ₄₂ indicates the introduction of two conjugate complex numbers, and 3 in C ₆₃ indicates the introduction of three Conjugate plurals. 4. The method according to claim 1, wherein, for each of the signals to be observed, the energy feature and the cumulant feature of the signal to be observed are used as input data, and the pre-trained The support vector machine SVM classifier utilizes the classification decision function in the SVM classifier to determine whether the unmanned aerial vehicle signal is included in the signal to be detected, including: For each of the signals to be observed, the energy feature and the cumulant feature of the signal to be observed are used as vector elements to generate a corresponding support vector, wherein the support vector corresponding to the nth signal to be observed Expressed as: _{En represents the energy feature of the signal to be observed, γ n1 represents} the first cumulant feature of the signal to be observed, and γ _n2 represents the second cumulant feature of the observed signal; Using the support vectors as input data into a pre-trained SVM classifier; According to the classification decision function in the SVM classifier, determine the value corresponding to the signal to be observed as a decision value, wherein the classification decision function f(m _n ) is expressed as: The value of the classification decision function f(m _n ) represents the decision value corresponding to the nth signal to be observed, sign( ) represents a sign function, and N represents the number of sample sub-signals, Indicates the dual value of the Lagrangian multiplier corresponding to the i-th sample support vector, <·> indicates the inner product operation, b ^* indicates the value of the bias, m _n ^T indicates the energy of the n-th signal to be observed The transposition of the support vector corresponding to the feature and the cumulant feature, x _i represents the energy feature of the i-th sample sub-signal and the sample support vector corresponding to the cumulant feature, and y _i represents the label value corresponding to the i-th sample sub-signal; Based on the decision value of each signal to be observed, it is determined whether the signal to be detected contains a drone signal, and when at least one of the decision values is 1, it means that the signal to be detected is UAV signals are included, and when the decision values are all -1, it means that the signals to be detected do not include UAV signals. 5. The method according to claim 1, the training process of the SVM classifier, comprising: Obtain a preset number of wireless signals in the frequency band to be detected according to preset conditions as sample signals; dividing the sample signal according to a preset frequency interval to obtain wireless sub-signals in multiple frequency bands as sample sub-signals; For each of the sample sub-signals, extracting energy features and cumulant features of the sample sub-signals; Based on the preset number of sample signals, the sample support vector composed of the energy feature and cumulant feature of the sample sub-signal in the sample signal, and the label value corresponding to the sample sub-signal are used as input data, and the input adopts the current classification The SVM classifier with parameters and a preset structure uses the current classification decision function to classify the preset number of sample signals to complete a round of training, wherein, during the first training, the parameters of the current classifier are preset initial classifier parameters, the current classification decision function is determined according to the preset initial classifier parameters; For each round of training, based on the preset number of sample signals, according to a preset loss function, determine the loss value of the SVM classifier; When it is determined based on the loss value that the SVM classifier reaches a preset standard, the training is completed, and the classification decision function corresponding to the SVM classifier is determined; When it is determined based on the loss value that the SVM classifier does not meet the preset standard, adjust the classifier parameters according to the preset adjustment method to obtain a new classification decision function, and adopt the new classification decision function and the preset Number of sample signals to complete a new round of training. 6. The method according to claim 5, wherein, for each round of training, based on the preset number of sample signals, according to a preset loss function, determining the loss value of the SVM classifier includes : After each round of training is completed, based on the hinge loss function, the SVM classifier loss value is determined, wherein the hinge loss function P(x) is expressed as: The value of the hinge loss function P(x) represents the loss value, λ represents the parameter to be adjusted, w represents the normal vector of the segmentation plane in the SVM classifier, ||·|| represents the norm, and N represents the The number of sample sub-signals described above, w ^T represents the transpose of w, x _i represents the sample support vector corresponding to the energy feature and cumulant feature of the i-th sample sub-signal, y _i represents the label corresponding to the i-th sample sub-signal value, and b represents the intercept. 7. The method according to claim 5, wherein, when determining that the SVM classifier does not reach a preset standard based on the loss value, adjust the classifier parameters according to a preset adjustment method to obtain a new The classification decision function, and use the new classification decision function and the preset number of sample signals to complete a new round of training, including: When it is determined based on the loss value that the SVM classifier does not meet the preset standard, based on the hinge loss function, using a gradient descent method to determine the value of the parameter to be adjusted in the hinge loss function; Re-determine the classification decision function corresponding to the SVM classifier according to the value of the parameter to be adjusted; Classify the preset number of sample signals according to the re-determined classification decision function to complete a new round of training. 8. The method according to claim 7, wherein the re-determining the classification decision function corresponding to the SVM classifier according to the value of the parameter to be adjusted comprises: According to the value of the parameter to be adjusted, determine the penalty parameter of the SVM classifier, wherein the penalty parameter represents the weight of adjusting the function interval and classification accuracy in the optimization direction, the penalty parameter D and the parameter to be adjusted The relationship between λ is expressed as: Based on the penalty parameter, using a convex optimization problem, determine the division method corresponding to the segmentation surface of the SVM classifier, wherein the division method of the segmentation surface represents a method for classifying the preset number of sample signals using a hyperplane , specifically expressed as: sty _i (w ^T x _i +b)≥1-ξ _i , i=1, 2,..., N ξ _i ≥ 0, i = 1, 2, ..., N st represents the constraint condition, w represents the normal vector of the hyperplane, ||·|| represents the norm, D represents the penalty parameter, N represents the number of the sample sub-signals, ξ _i represents the i-th sample sub-signal corresponding The slack variable of , y _i (w ^T x _i +b) represents the interval distance from the i-th sample support vector to the segmentation plane, y _i represents the label value corresponding to the i-th sample sub-signal, w ^T represents the transposition of w, x _i represents the i-th sample support vector, b represents the intercept; Based on the Lagrangian function and the function dualization method, the solution problem of the division method is converted, and the division method after conversion is expressed as: 0≤α _i ≤D, i=1, 2,...,N st represents the constraints, α _i represents the value of the Lagrange multiplier corresponding to the i-th sample sub-signal, α _j represents the value of the Lagrange multiplier corresponding to the j-th sample sub-signal, and y _i represents the value of the i-th sample sub-signal The label value corresponding to the sample sub-signal, y _j represents the label value corresponding to the i-th sample sub-signal, The transpose of the sample support vector corresponding to the i-th sample sub-signal, x _j represents the sample support vector corresponding to the j-th sample sub-signal, <·> represents the inner product operation, D represents the penalty parameter, and N represents the number of sample sub-signals ; Determine the value of the Lagrange multiplier based on the SMO algorithm; Utilize the properties of the sample support vector to determine the value of the bias, wherein the value b ^* of the bias is expressed as: N represents the number of sample sub-signals, y _n represents the label value corresponding to the n-th sample support vector, y _i represents the label value corresponding to the i-th sample support vector, Indicates the dual value of the Lagrangian multiplier corresponding to the i-th sample support vector, Indicates the transpose of the i-th sample support vector, x _n indicates the sample support vector corresponding to the n-th sample sub-signal, <·> indicates the inner product operation; Based on the value of the Lagrangian multiplier and the value of the bias, the classification decision function is re-determined. 9. A UAV signal detection device, characterized in that, comprising: The signal acquisition module to be detected is used to obtain the wireless signal in the frequency band to be detected as the signal to be detected; The signal to be observed acquisition module is used to divide the signal to be detected according to the preset frequency interval, and obtain wireless sub-signals in multiple frequency bands as the signal to be observed; An energy feature extraction module, configured to extract the energy feature of the signal to be observed based on a preset energy feature formula for each signal to be observed; A cumulant feature extraction module, configured to extract the cumulant feature of the signal to be observed based on a preset cumulant feature formula for each of the signals to be observed; The unmanned aerial vehicle signal determination module is used for each of the signals to be observed, using the energy feature and the cumulative feature of the signal to be observed as input data, inputting a pre-trained SVM classifier, using the SVM The classification decision function in the classifier determines whether the unmanned aerial vehicle signal is included in the signal to be detected, wherein the SVM classifier is trained based on the feature data of the sample signal, and the feature data of the sample signal includes A sample support vector formed by the energy feature and cumulant feature of each sample sub-signal in the sample signal, and a flag value corresponding to the sample sub-signal, the flag value indicating whether there is no one in the sample sub-signal machine signal. 10. The device according to claim 9, further comprising: The sample signal acquisition module is used to acquire a preset number of wireless signals in the frequency band to be detected as sample signals according to preset conditions; A sample sub-signal acquisition module, configured to divide the sample signal according to a preset frequency interval, and obtain wireless sub-signals in multiple frequency bands as sample sub-signals; A feature data acquisition module, configured to extract energy features and cumulant features of the sample sub-signal for each of the sample sub-signals; A sample signal classification module, configured to use, based on the preset number of sample signals, a sample support vector composed of energy features and cumulant features of sample sub-signals in the sample signal, and a marker value corresponding to the sample sub-signal as Input data, input an SVM classifier using current classifier parameters and a preset structure, use the current classification decision function to classify the preset number of sample signals, and complete a round of training, wherein, during the first training, the said The current classifier parameters are preset initial classifier parameters, and the current classification decision function is determined according to the preset initial classifier parameters; A loss value determination module, configured to determine the loss value of the SVM classifier according to a preset loss function based on the preset number of sample signals for each round of training; A classifier generating module, configured to complete training when determining that the SVM classifier reaches a preset standard based on the loss value, and determine a classification decision function corresponding to the SVM classifier; A classifier parameter adjustment module, configured to adjust the classifier parameters according to a preset adjustment method when it is determined based on the loss value that the SVM classifier does not reach a preset standard, to obtain a new classification decision function, and adopt a new Classify the decision function and the preset number of sample signals to complete a new round of training.