UAV communication jamming task generation method based on deep learning

Abstract

The present invention relates to the technical field of communication signal processing, and proposes a method for generating a UAV communication jamming task based on deep learning, including: obtaining a communication mixed signal; obtaining a component feature matching vector according to a spectrum diagram of the communication mixed signal; obtaining a local amplitude sequence according to the slope information at each frequency; obtaining a component feature matching degree according to the local amplitude sequence; obtaining a segmented area in a frequency feature line graph using a threshold segmentation algorithm; obtaining a sub-signal decomposition result according to the number of segmented areas using an independent component analysis algorithm; obtaining a weak interference signal according to the sub-signal decomposition result using a waveform matching algorithm; generating an interference counter signal according to the weak interference signal using a deep learning model, and completing the jamming of the UAV communication based on the interference counter signal. The present invention obtains the number of components in the independent component analysis algorithm using the distribution characteristics of the spectrum diagram of the communication mixed signal, thereby improving the separation accuracy of the independent component analysis algorithm for the communication mixed signal.

Description

Technical Field

[0001] The present invention relates to the technical field of communication signal processing, and in particular to a method for generating unmanned aerial vehicle communication jamming tasks based on deep learning. Background Art

[0002] UAV, also known as unmanned aircraft, is an aircraft controlled by a ground remote control device and an onboard program. It is used in many fields such as aerial photography and mapping, agricultural plant protection, and express delivery. During the flight of a UAV, the UAV may mistakenly enter sensitive areas such as unauthorized entry and no-fly zones, or even human-controlled UAVs may intrude into sensitive areas. Therefore, it is necessary to design an active defensive UAV countermeasure method to counter and counter the communication interference caused by the UAV's communication system to the communication system in sensitive areas, prevent illegal intrusion of UAVs, and maintain public safety and personal privacy.

[0003] In the process of active defense drone countermeasures, drones are interference sources with a constantly changing relative spatial distance to the communication system in sensitive areas. The countermeasure target of the communication system in sensitive areas is the drone communication signal generated by the drone communication system. However, with the development of communication technology, drones themselves detect external interference by equipping multiple sensors to improve the stability of their own communications. Therefore, it is necessary to analyze the anti-interference ability of drones, distinguish the strengths and weaknesses of drone anti-interference capabilities, and use deep learning technology to generate targeted anti-interference countermeasure signals for weak anti-interference in drone communications, so as to achieve communication countermeasure interference for drones that break into sensitive areas, or send expulsion signals to drones to eliminate security risks in sensitive areas. However, due to the instability of the spatial position of drones, the directional differences of various sub-signals in the communication mixed signals received by the receiving end of the communication system in sensitive areas are reduced, and the correlation is increased. Therefore, it is necessary to accurately decompose the received drone communication signals in the process of identifying the strengths and weaknesses of drone anti-interference capabilities. Summary of the invention

[0004] The present invention provides a method for generating UAV communication jamming tasks based on deep learning to solve the problem that the traditional signal decomposition algorithm is difficult to effectively separate the UAV communication signal at the signal receiving end. The technical solution adopted is as follows:

[0005] An embodiment of the present invention provides a method for generating a UAV communication jamming task based on deep learning, the method comprising the following steps:

[0006] Acquire the mixed communication signal generated by the UAV communication system;

[0007] Obtain a component feature matching vector of each frequency in the communication mixed signal spectrum diagram according to the time domain information of the communication mixed signal; obtain a local amplitude sequence of each frequency according to the slope information at each frequency in the communication mixed signal spectrum diagram; obtain a component feature matching degree of each frequency according to the local amplitude sequence of each frequency;

[0008] A threshold segmentation algorithm is used to obtain the segmented areas in the frequency feature line graph constructed by the component feature matching degree of all frequencies; an independent component analysis algorithm is used to obtain the sub-signal decomposition result of the communication mixed signal based on the number of the segmented areas;

[0009] According to the sub-signal decomposition results of the communication mixed signal, the waveform matching algorithm is used to obtain the weak interference signal of the UAV communication; according to the weak interference signal of the UAV communication, the deep learning model is used to generate the interference counter-signal acting on the UAV communication, and the interference to the UAV communication is achieved according to the interference counter-signal.

[0010] Preferably, the method for obtaining the component feature matching vector of each frequency in the spectrum diagram of the communication mixed signal according to the time domain information of the communication mixed signal is:

[0011] The Fourier transform result of the communication mixed signal is used as a spectrum diagram of the communication mixed signal;

[0012] The slope at each frequency in the spectrum diagram of the communication mixed signal is obtained respectively. For the frequency whose slope is greater than or equal to zero, the amplitude corresponding to the first wave peak determined along the direction in which the frequency increases is used as the reference amplitude of the frequency; for the frequency whose slope is less than zero, the amplitude corresponding to the first wave peak determined along the direction in which the frequency decreases is used as the reference amplitude of the frequency;

[0013] The vector composed of the amplitude, slope and reference amplitude of each frequency is used as the component feature matching vector of each frequency.

[0014] Preferably, the method for obtaining the local amplitude sequence of each frequency according to the slope information at each frequency in the communication mixed signal spectrum diagram is:

[0015] Each frequency of the communication mixed signal spectrum diagram is taken as the center frequency, and a preset number of amplitudes are obtained on the left and right sides of the center frequency respectively, and a sequence consisting of the preset number of amplitudes and the amplitude corresponding to the center frequency in ascending order of frequency is taken as the local amplitude sequence of each frequency.

[0016] Preferably, the method for obtaining the component feature matching degree of each frequency according to the local amplitude sequence of each frequency is:

[0017] The amplitude fluctuation index of each frequency is obtained according to the gradient change of adjacent elements in the local amplitude sequence of each frequency;

[0018] Obtain the component fitting coefficient of each frequency according to the fitting results of the elements in the local amplitude sequence of each frequency and the frequency corresponding to each frequency and the local amplitude extreme value;

[0019] The component feature matching degree of each frequency is composed of two parts: the amplitude fluctuation index of each frequency and the component fitting coefficient, wherein the component feature matching degree is proportional to the amplitude fluctuation index and the component fitting coefficient.

[0020] Preferably, the method for obtaining the amplitude fluctuation index of each frequency according to the gradient change of adjacent elements in the local amplitude sequence of each frequency is:

[0021] The sequence consisting of the remaining elements after deleting the last two elements from the local amplitude sequence of each frequency is used as the calculation sequence of each frequency, each element in the calculation sequence is marked as a calculation element, the next element of each calculation element is used as the first adjacent element of each calculation element, and the next element of the first adjacent element of each calculation element is used as the second adjacent element of each calculation element;

[0022] The difference between the first adjacent element of each calculation element and each calculation element is used as a denominator, the difference between the second adjacent element of each calculation element and the first adjacent element of each calculation element is used as a numerator, and the difference between the absolute value of the ratio of the numerator to the denominator and a preset parameter is used as an independent variable of the hyperbolic tangent function;

[0023] The function value corresponding to the independent variable of the hyperbolic tangent function is taken as the average of all calculation elements in the calculation sequence of each frequency as the amplitude fluctuation index of each frequency.

[0024] Preferably, the method of obtaining the component fitting coefficient of each frequency according to the fitting results of the elements in the local amplitude sequence of each frequency and the frequency corresponding to each frequency and the local amplitude extreme value is:

[0025] Obtaining the goodness of fit when a sine function is fitted to all elements in the local amplitude sequence of each frequency using the least squares method, and taking the inverse of the square of the goodness of fit as a proportional factor;

[0026] The Euclidean distance between the frequencies corresponding to the first trough and the second trough of each frequency is used as the denominator, and the ratio of each element in the component feature matching vector of each frequency to the denominator is accumulated on the component feature matching vector of each frequency as the first multiplication factor;

[0027] The normalized result of the product of the scale factor and the first multiplication factor is used as the component fitting coefficient for each frequency.

[0028] Preferably, the method of using the threshold segmentation algorithm to obtain the segmented area in the frequency feature line graph constructed by the component feature matching degree of all frequencies is:

[0029] The line connecting the data points in the two-dimensional coordinate system constructed with each frequency as the horizontal axis and the component feature matching degree corresponding to each frequency as the vertical axis is used as a frequency feature line graph;

[0030] The threshold segmentation algorithm is used to obtain the segmentation threshold of all data points in the frequency characteristic line graph, and the horizontal straight line with the segmentation threshold as the ordinate is used as the segmentation line in the frequency characteristic line graph;

[0031] The area formed by the local line segment composed of each continuous data point on the segmentation line and the segmentation line is taken as a segmentation area.

[0032] Preferably, the method of obtaining the sub-signal decomposition result of the communication mixed signal by using the independent component analysis algorithm based on the number of segmented regions is:

[0033] The number of segmented regions is used as the number of components in an independent component analysis algorithm, the communication mixed signal is used as the input of the independent component analysis algorithm, and the sub-signal decomposition result of the communication mixed signal is obtained by using the independent component analysis algorithm.

[0034] Preferably, the method of obtaining the weak interference signal of the UAV communication by using the waveform matching algorithm according to the sub-signal decomposition result of the communication mixed signal is:

[0035] The waveform matching algorithm is used to obtain the waveform matching degree between the sub-signal decomposition result of the communication mixed signal and the original interference signal, and the original interference signal corresponding to the maximum value of the waveform matching degree is used as the weak interference signal of the UAV communication.

[0036] Preferably, the method of generating an interference countermeasure signal acting on drone communication using a deep learning model according to a weak interference signal of drone communication is:

[0037] The weak interference signal of UAV communication is used as the input of the generative adversarial network, and the output of the generative adversarial network is used as the interference countermeasure signal of UAV communication.

[0038] The beneficial effects of the present invention are as follows: the present invention constructs a component feature matching vector by analyzing the features of the UAV communication signal in the communication mixed signal spectrum diagram after being interfered with, and constructs an amplitude fluctuation index based on the component feature matching vector, reflecting the variation characteristics of the UAV communication signal after being converted into a spectrum; secondly, the component feature matching degree is constructed by combining the component feature matching vector and the amplitude fluctuation index, and the matching degree of each frequency belonging to the component feature in the communication mixed signal spectrum diagram is comprehensively considered; the number of components in the ICA algorithm is obtained based on the segmentation area result of the frequency feature line graph, which solves the problem of low accuracy of the traditional ICA algorithm in separating the UAV communication mixed signal; the sub-signal decomposition result of the communication mixed signal is obtained based on the improved ICA algorithm, and finally the generative adversarial network is used to adaptively generate interference signals to the UAV, so as to complete the interference tasks for different UAV communications. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 A schematic diagram of a flow chart of a method for generating a UAV communication jamming task based on deep learning provided by one embodiment of the present invention;

[0041] Figure 2 A schematic diagram of a frequency characteristic line graph provided by an embodiment. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] See also Figure 1 , which shows a flow chart of a method for generating a UAV communication jamming task based on deep learning provided by an embodiment of the present invention, the method comprising the following steps:

[0044] Step S001, obtaining a mixed communication signal generated by the UAV communication system.

[0045] Since the countermeasures against drones in sensitive areas are proactively defensive, the communication system in the sensitive areas actively sends different types of interference signals intermittently to drones detected within the communication range of the sensitive areas in the present invention. Secondly, the receiving end of the communication system in the sensitive areas receives the mixed communication signals returned by the drones. Since the strength of the communication signal gradually weakens during the communication process, the differences in different directions gradually decrease. Therefore, in order to generate a communication interference task for anti-interference of the drone's weaknesses, it is first necessary to accurately decompose the drone's weak anti-interference signals from the communication mixed signals. In the present invention, the specific process of the mixed communication signals received by the communication system in the sensitive areas after each communication is as follows:

[0046] When the ground communication center detects the presence of a drone within the communication range, it first uses the photoelectric detection module to obtain the flight trajectory of the drone, and then sends different interference signals to the drone at fixed communication time intervals, and obtains the drone communication information data through the communication receiver. The drone communication information data includes the drone's appearance image, the altitude information on the drone's flight trajectory obtained by the external radiation source radar, the longitude and latitude information, and the flight speed information.

[0047] Since the data collection process of UAV communication information may be affected by natural noise, channel fading and other factors, the collected data may have abnormal conditions such as noise. Therefore, the present invention performs denoising on the collected UAV communication signal data. Common denoising technologies include digital filtering, adaptive filtering, wavelet transform, etc. In order to adapt to the changes of different signals and noises and achieve better denoising effects, the present invention uses adaptive filtering technology to perform denoising on the collected UAV communication signal data. Adaptive filtering technology is a well-known technology, and the specific process will not be repeated in this invention. The signal obtained by modulating the pre-processed UAV communication signal data is used as a communication mixed signal. Signal modulation is a well-known technology, and the specific process will not be repeated in this invention.

[0048] At this point, the communication mixed signal received by the ground communication center after each communication is obtained.

[0049] Step S002, obtaining the component feature matching degree of each frequency based on the component feature matching vector of each frequency and the local amplitude sequence.

[0050] The purpose of the present invention is to counter the communication interference generated by drones entering sensitive areas. The implementation method of the countermeasure method is to use deep learning technology to generate interference countermeasure signals for weak anti-interference in drone communications, and use the interference countermeasure signals to achieve interference and countermeasures in drone communications, avoiding the potential safety hazards caused by drones to sensitive areas. In the current active defensive drone countermeasure methods, the decomposition of weak anti-interference signals in drone communications mostly uses independent component analysis algorithms. However, since the independent component analysis ICA algorithm calculates more statistical features of the data, in order to reduce the amount of algorithm calculations and improve the calculation efficiency, and since a variety of noise interferences are emitted to the drone, the drone communication signal contains more signal types, that is, the obtained drone communication signal is a mixed signal. Based on this, the present invention needs to perform centralization and whitening on the mixed signal before using the ICA algorithm to separate the obtained communication mixed signal.

[0051] The centralization process is to remove the mean value, so that the mean value of the mixed signal is zero, thereby reducing the amount of calculation of the ICA algorithm, specifically by making a difference between the original mixed signal and its mathematical expectation to obtain a mixed signal after centering; the whitening process is to remove the correlation between the mixed signals, which can simplify the process of extracting independent components by the ICA algorithm, thereby improving the operational efficiency of the ICA algorithm, specifically by solving the whitening matrix by the principal component analysis method, performing matrix multiplication of the obtained whitening matrix and the mixed signal after centering, and obtaining a whitening vector, wherein each element in the vector is each component in the mixed signal. The principal component analysis method is a known technology, and the specific process of the present invention will not be repeated.

[0052] Most modern communication systems usually use digital modulation technology, and the UAV communication system also uses digital modulation technology, which includes QPSK, QAM, OFDM, etc. These signals are usually considered to be non-Gaussian signals, and have complex modulation structures during transmission, and UAV communication may be affected by a variety of interference signals, the properties of which may be different, that is, each interference signal can be considered to be independent of each other and non-Gaussian signal. Since non-Gaussian signals have more nonlinear properties, that is, there may be greater differences between different components, and the probability distribution of non-Gaussian signals is usually not a symmetrical curve, but a more complex distribution, and contains more information, the effect of using the ICA algorithm to separate them is better, and accordingly, the present invention performs Fourier transform on the communication mixed signal, and the spectrum corresponding to the communication mixed signal is recorded as the communication mixed signal spectrum. Wherein Fourier transform is a well-known technology, and the specific process of the present invention will not be repeated.

[0053] Based on the slope information at each frequency in the spectrum of the communication mixed signal, a component feature matching vector of each frequency is constructed, and the component feature matching vector at frequency f is recorded as HV f :

[0054] HV f =(A f ,S f ,H f )

[0055] Among them, HV f Represents the component feature matching vector at frequency f in the spectrum of the communication mixed signal, where A f It represents the amplitude of the communication mixed signal at frequency f in the spectrum diagram, S f It represents the slope of the frequency f in the spectrum of the communication mixed signal. f ≥0, the amplitude corresponding to the first peak determined along the direction of increasing frequency is taken as the reference amplitude of frequency f. f <0, the amplitude corresponding to the first peak determined along the direction of decreasing frequency is taken as the reference amplitude of frequency f, H f Represents the reference amplitude of frequency f in the mixed signal spectrum.

[0056] On the other hand, since the UAV communication signal may be interfered with, the peak in the communication mixed signal spectrum may be weakened, a new peak may appear, and the communication mixed signal spectrum may be distorted, resulting in inaccurate estimation of the number of components of the ICA algorithm. Therefore, the present invention constructs a component feature matching degree based on the component feature matching vector obtained in the above steps. Taking frequency f as an example, the process of obtaining the component feature matching degree of frequency f is as follows:

[0057] The first trough found in the communication mixed signal spectrum along the direction of decreasing frequency is recorded as the first trough of frequency f, and the first trough found in the communication mixed signal spectrum along the direction of increasing frequency is recorded as the second trough of frequency f. The frequency f is taken as the center frequency. The amplitudes corresponding to k adjacent frequencies are taken on the left and right sides of the communication mixed signal spectrum. The amplitudes corresponding to the k adjacent frequencies on the left and right sides and the amplitudes corresponding to frequency f are taken in ascending order of frequency as the local amplitude sequence W(f) of frequency f. In the present invention, the value of k is taken as an empirical value of 10. Secondly, all elements in the sequence W(f) are fitted with a sine function using the least squares method, and the goodness of fit of the obtained fitting result is recorded as r. f The least square method is a well-known technology, and the specific process will not be described in detail in this invention.

[0058] Calculate the component characteristic matching degree HVD of frequency f f :

[0059]

[0060]

[0061] In the formula, ACFf is the amplitude fluctuation index of frequency f, w is the number of elements in the local amplitude sequence W(f), tanh() is the hyperbolic tangent function, W f,j , W f,j+1 , W f,j+2 The jth, j+1th, and j+2th elements of the local amplitude sequence W(f) of frequency f, respectively;

[0062] Norm[] is the normalization function, m is the number of elements in the component feature matching vector, the value of m is 3, D f It is the Euclidean distance between the frequency corresponding to the first trough of frequency f and the frequency corresponding to the second trough in the spectrum diagram of the communication mixed signal.

[0063] Since the ground communication center sends a variety of interference signals to the drone, the curve shape in the communication mixed signal spectrum is distorted to a large extent, making the originally smooth spectrum show irregular fluctuations. Therefore, the stronger the difference between the adjacent elements in the local amplitude sequence of the frequency f in the communication mixed signal spectrum, that is, The greater the difference from 1, the greater the The larger the value of is, the larger the absolute value of the amplitude fluctuation index obtained after the hyperbolic tangent function transformation is. f The larger the value of , the greater the calculated component feature matching degree; the smaller the Euclidean distance between the frequency of the first trough of frequency f and the frequency of the second trough of frequency f, the larger the amplitude of frequency f, and the larger the reference amplitude of frequency f, that is, D f The smaller the value, the smaller the HV f The larger the value of (i), the sharper the peak shape. The larger the value, the weaker the predictability of the spectrum data in the sequence and the more irregular the spectrum changes. The larger the value of The larger the value, the HVD f The larger the value of , the more the communication mixed signal is concentrated in a smaller frequency range, the more obvious the component characteristics are, and the greater the calculated component characteristic matching degree should be.

[0064] At this point, the component feature matching degree of each frequency in the mixed signal spectrum is obtained, which is convenient for determining the number of components in the subsequent independent component analysis algorithm.

[0065] Step S003, using a threshold segmentation algorithm to obtain segmented regions in the frequency characteristic line graph, and using an independent component analysis algorithm based on the number of segmented regions to obtain a sub-signal decomposition result of the communication mixed signal.

[0066] In the process of actively defending against drone communications, the present invention improves the decomposition algorithm of the communication mixed signal in the process of generating the drone communication interference task, and accurately identifies the weak anti-interference in the drone communication based on the improved independent component analysis algorithm, which facilitates the subsequent deep learning model training.

[0067] In order to achieve countermeasures against drone communications, the above content obtains the component feature matching degree of each frequency through the feature matching vector in the communication mixed signal. It is also necessary to further obtain the key parameters in the independent component analysis algorithm, that is, the number of components in the independent component analysis algorithm, and consider further analyzing the differences in the decomposable sub-signals in the communication mixed signal according to the component feature matching degree, to determine the number of components and complete the improvement of the independent component analysis.

[0068] According to the component feature matching degree of each frequency in the communication mixed signal spectrum obtained in the above steps, the line connecting the data points in the two-dimensional coordinate system constructed with each frequency as the horizontal axis and the component feature matching degree corresponding to each frequency as the vertical axis is used as the frequency feature line graph, such as Figure 2 As shown. Secondly, the Otsu threshold segmentation method is used to obtain the matching degree threshold THR of all component features, and the obtained threshold is used as the dividing line of the frequency feature line graph, and the area composed of the local line segment composed of each continuous data point on the dividing line and the dividing line is used as a segmentation area, and the number of all segmentation areas in the frequency feature line graph is used as the number of components, and the obtained number of components is recorded as K. The Otsu threshold segmentation method is a well-known technology, and the specific process will not be repeated in this invention.

[0069] The number of components K obtained through the above steps is used as the number of components of the ICA algorithm, and the communication mixed signal is used as the input of the ICA algorithm. The decomposed independent signal vector can be obtained through the ICA algorithm. The length of the obtained vector is K, and each element in the vector is an independent signal. The decomposed independent signal is recorded as a UAV communication sub-signal. The obtained UAV communication sub-signal contains the original interference signal and the effective communication signal emitted. The ICA algorithm is a well-known technology, and the specific process will not be repeated here.

[0070] At this point, the sub-signal decomposition results of the communication mixed signal are obtained, which are used for the subsequent deep learning model to generate interference suppression signals.

[0071] Step S004, using a waveform matching algorithm to obtain a weak interference signal of the UAV communication, obtaining an interference counter signal based on a generative adversarial network, and suppressing the UAV communication according to the interference counter signal.

[0072] In the process of using deep learning technology to generate interference countermeasure signals for weak anti-interference in drone communications, the present invention focuses on the decomposition process of sub-signals in communication mixed signals received in communication systems in sensitive areas, and performs similarity measurement with existing drone original interference signals on the basis of sub-signal decomposition. Based on the similarity measurement, a generative adversarial network is used to obtain interference countermeasure signals for countering drones. Communication systems in sensitive areas use interference countermeasure signals to achieve defensive countermeasures against interference in drone communications, thereby ensuring that communication systems in sensitive areas are protected from interference in drone communications.

[0073] According to the above steps, the sub-signal decomposition result of the communication mixed signal is obtained, and then the waveform matching algorithm is used to calculate the waveform matching degree between each drone communication sub-signal and each original interference signal. The waveform matching algorithm is a well-known technology and the specific process is not repeated. For example, the greater the waveform matching degree between the drone communication sub-signal and a class of original interference signals, the higher the matching degree between the drone communication sub-signal and a class of original interference signals, that is, the drone communication signal contains more interference signals of this category, indicating that the drone has a weaker anti-interference ability to this category of interference signals; the smaller the waveform matching degree between the drone communication sub-signal and a class of original interference signals, the lower the matching degree between the drone communication sub-signal and a class of original interference signals, that is, the drone communication signal contains fewer such interference signals, indicating that the drone has a stronger anti-interference ability to the interference signal.

[0074] Furthermore, a type of original interference signal corresponding to the maximum value of all waveform matching degrees is used as the weak anti-interference signal of the UAV. Secondly, the weak anti-interference signal of the UAV is used as the input of the generative adversarial network GAN. The loss function of the generative adversarial network includes the loss function of the discriminator D and the loss function of the generator G. Among them, the role of the discriminator is to identify whether the sample input into the discriminator is a real sample or a sample generated by the generator. Therefore, the loss function of the discriminator is set to the binary cross entropy function; and the role of the generator is to make the generated samples close to the real samples, so the loss function of the generator adopts the mean square error loss function.

[0075] The sum of the loss function of the discriminator and the loss function of the generator is used as the loss function of the generative adversarial network. The Adam algorithm is used as the optimization algorithm, and the output of the generative adversarial network is used as the interference countermeasure signal. The GAN network is supervised learning, so the output of the generative adversarial network, that is, the interference countermeasure signal output by the generator needs to be manually labeled. Specifically, experts can manually set the corresponding interference countermeasure signal according to the weak signals of different drones. Specifically, firstly, the weak interference signal database of existing common drones is established using the expert prior knowledge, and the matching degree of the interference countermeasure signal obtained by the waveform matching algorithm and the weak interference signal of each type of drone in the weak interference signal database is recorded as the label value of the interference countermeasure signal. Secondly, the Otsu threshold segmentation algorithm is used to obtain the segmentation threshold of all the label values ​​of the interference countermeasure signal, and the set of weak interference signals corresponding to all label values ​​greater than the segmentation threshold is used as the candidate signal set of the interference countermeasure signal. It is a well-known technology, and the specific process is not repeated.

[0076] Further, the mean of the energy peaks of all weak interference signals in the candidate signal set is calculated respectively, and the mean of the energy peaks is compared with the energy mean in the obtained communication mixed signal. If the mean of the energy peaks is greater than the energy mean in the communication mixed signal, it is considered that the radio signal in the UAV communication process is strong, and the communication system in the sensitive area uses a directional radio frequency and interference countermeasure signal with a certain power to send to the UAV, so that the UAV cannot receive specific coordinate information, and it loses control or falls; if the mean of the energy peaks is less than the energy mean in the communication mixed signal, it is considered that the radio signal in the UAV communication process is weak, and the frequency band with the highest probability of appearing in the interference countermeasure signal is obtained at this time. The communication system in the sensitive area blocks the frequency band with the highest probability of appearing in the interference countermeasure signal by blocking the interference of the remote control signal frequency band, so as to achieve countermeasure against the UAV. The specific process of constructing a database with expert prior knowledge is well known to those skilled in the art, including the training of the generative adversarial network, which is a well-known technology, and the specific process will not be repeated. Since the input of the generative adversarial network is the weak anti-interference signal of the UAV, the UAV also has weak anti-interference ability to the obtained interference countermeasure signal. The interference countermeasure signal output by the generative adversarial network is sent to the UAV to complete the adaptive weak anti-interference generation task for the UAV.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for generating UAV communication jamming tasks based on deep learning, characterized in that: The method comprises the following steps: Acquire the mixed communication signal generated by the UAV communication system; the mixed communication signal is obtained by filtering and denoising the UAV communication signal data, and the UAV communication signal includes the UAV appearance image, the height information, longitude and latitude information and flight speed information on the UAV flight trajectory obtained by the external radiation source radar; Obtain a component feature matching vector of each frequency in the communication mixed signal spectrum diagram according to the time domain information of the communication mixed signal; obtain a local amplitude sequence of each frequency according to the slope information at each frequency in the communication mixed signal spectrum diagram; obtain a component feature matching degree of each frequency according to the local amplitude sequence of each frequency; A threshold segmentation algorithm is used to obtain the segmented areas in the frequency feature line graph constructed by the component feature matching degree of all frequencies; an independent component analysis algorithm is used to obtain the sub-signal decomposition result of the communication mixed signal based on the number of the segmented areas; According to the sub-signal decomposition results of the communication mixed signal, the waveform matching algorithm is used to obtain the weak interference signal of the UAV communication; according to the weak interference signal of the UAV communication, the deep learning model is used to generate the interference counter signal acting on the UAV communication, and the interference to the UAV communication is realized according to the interference counter signal; The method for obtaining the component feature matching vector of each frequency in the spectrum diagram of the communication mixed signal according to the time domain information of the communication mixed signal is: The Fourier transform result of the communication mixed signal is used as a spectrum diagram of the communication mixed signal; The slope at each frequency in the spectrum diagram of the communication mixed signal is obtained respectively. For the frequency whose slope is greater than or equal to zero, the amplitude corresponding to the first wave peak determined along the direction in which the frequency increases is used as the reference amplitude of the frequency; for the frequency whose slope is less than zero, the amplitude corresponding to the first wave peak determined along the direction in which the frequency decreases is used as the reference amplitude of the frequency; The vector composed of the amplitude, slope and reference amplitude of each frequency is used as the component feature matching vector of each frequency; The method for obtaining the component feature matching degree of each frequency according to the local amplitude sequence of each frequency is: The amplitude fluctuation index of each frequency is obtained according to the gradient change of adjacent elements in the local amplitude sequence of each frequency; Obtain the component fitting coefficient of each frequency according to the fitting results of the elements in the local amplitude sequence of each frequency and the frequency corresponding to each frequency and the local amplitude extreme value; The component feature matching degree of each frequency is composed of two parts: the amplitude fluctuation index of each frequency and the component fitting coefficient, wherein the component feature matching degree is proportional to the amplitude fluctuation index and the component fitting coefficient; The method for obtaining the component fitting coefficient of each frequency according to the fitting results of the elements in the local amplitude sequence of each frequency and the frequency corresponding to the local amplitude extreme value of each frequency is: Obtaining the goodness of fit when a sine function is fitted to all elements in the local amplitude sequence of each frequency using the least squares method, and taking the inverse of the square of the goodness of fit as a proportional factor; The Euclidean distance between the frequencies corresponding to the first trough and the second trough of each frequency is used as the denominator, and the ratio of each element in the component feature matching vector of each frequency to the denominator is accumulated on the component feature matching vector of each frequency as the first multiplication factor; The normalized result of the product of the scale factor and the first multiplication factor is used as the component fitting coefficient of each frequency; The method of using the threshold segmentation algorithm to obtain the segmented area in the frequency feature line graph constructed by the component feature matching degree of all frequencies is: The line connecting the data points in the two-dimensional coordinate system constructed with each frequency as the horizontal axis and the component feature matching degree corresponding to each frequency as the vertical axis is used as a frequency feature line graph; The threshold segmentation algorithm is used to obtain the segmentation threshold of all data points in the frequency characteristic line graph, and the horizontal straight line with the segmentation threshold as the ordinate is used as the segmentation line in the frequency characteristic line graph; The area formed by the local line segment composed of each continuous data point on the segmentation line and the segmentation line is taken as a segmentation area.

2. The method for generating UAV communication jamming tasks based on deep learning according to claim 1 is characterized in that: The method for obtaining the local amplitude sequence of each frequency according to the slope information at each frequency in the spectrum diagram of the communication mixed signal is: Each frequency of the communication mixed signal spectrum diagram is taken as the center frequency, and a preset number of amplitudes are obtained on the left and right sides of the center frequency respectively, and a sequence consisting of the preset number of amplitudes and the amplitude corresponding to the center frequency in ascending order of frequency is taken as the local amplitude sequence of each frequency.

3. The method for generating UAV communication jamming tasks based on deep learning according to claim 1 is characterized in that: The method for obtaining the amplitude fluctuation index of each frequency according to the gradient change of adjacent elements in the local amplitude sequence of each frequency is: The sequence consisting of the remaining elements after deleting the last two elements from the local amplitude sequence of each frequency is used as the calculation sequence of each frequency, each element in the calculation sequence is marked as a calculation element, the next element of each calculation element is used as the first adjacent element of each calculation element, and the next element of the first adjacent element of each calculation element is used as the second adjacent element of each calculation element; The difference between the first adjacent element of each calculation element and each calculation element is used as a denominator, the difference between the second adjacent element of each calculation element and the first adjacent element of each calculation element is used as a numerator, and the difference between the absolute value of the ratio of the numerator to the denominator and a preset parameter is used as an independent variable of the hyperbolic tangent function; The function value corresponding to the independent variable of the hyperbolic tangent function is taken as the average of all calculation elements in the calculation sequence of each frequency as the amplitude fluctuation index of each frequency.

4. The method for generating UAV communication jamming tasks based on deep learning according to claim 1 is characterized in that: The method for obtaining the sub-signal decomposition result of the communication mixed signal by using the independent component analysis algorithm based on the number of segmented regions is: The number of segmented regions is used as the number of components in an independent component analysis algorithm, the communication mixed signal is used as the input of the independent component analysis algorithm, and the sub-signal decomposition result of the communication mixed signal is obtained by using the independent component analysis algorithm.

5. The method for generating UAV communication jamming tasks based on deep learning according to claim 1 is characterized in that: The method for obtaining the weak interference signal of UAV communication by using the waveform matching algorithm according to the sub-signal decomposition result of the communication mixed signal is: The waveform matching algorithm is used to obtain the waveform matching degree between the sub-signal decomposition result of the communication mixed signal and the original interference signal, and the original interference signal corresponding to the maximum value of the waveform matching degree is used as the weak interference signal of the UAV communication.

6. The method for generating UAV communication jamming tasks based on deep learning according to claim 1 is characterized in that: The method for generating an interference countermeasure signal acting on drone communication by using a deep learning model according to the weak interference signal of drone communication is: The weak interference signal of UAV communication is used as the input of the generative adversarial network, and the output of the generative adversarial network is used as the interference countermeasure signal of UAV communication.

Images