An unmanned aerial vehicle unsupervised clustering detection method and system based on radio frequency signals

Abstract

The application provides a kind of unmanned aerial vehicle unsupervised clustering detection method based on radio frequency signal, comprising: collecting unmanned aerial vehicle radio frequency signal, and sampling unmanned aerial vehicle signal time-frequency spectrogram and unmanned aerial vehicle signal time-frequency azimuth chart, and merging to form time-frequency azimuth matrix;With two-dimensional binning to the time-frequency azimuth matrix denoising;The time-frequency azimuth matrix after denoising is standardized, and the clustering model is trained, and each frequency point is classified using the clustering model;According to the classification result, time-azimuth-intensity chart is generated;Each time-azimuth-intensity chart is denoised using a sliding window, according to the motion properties of the unmanned aerial vehicle target, the static non-unmanned aerial vehicle target is removed, the target azimuth information is extracted, and the target detection is completed.The application improves the unmanned aerial vehicle target detection ability in complex electromagnetic environment by automatic denoising based on entropy and real-time multi-dimensional clustering processing without increasing processing cost and limited complexity.

Description

Technical Field

[0001] This invention relates to the field of signal detection, and in particular to an unsupervised clustering detection method and system for unmanned aerial vehicles based on radio frequency signals. Background Technology

[0002] Existing drone detection methods include radar-based target detection, visual imaging-based detection, sensor-based detection, and radio frequency (RF) signal-based detection. Among these, RF signal-based drone detection methods use an RF signal receiver to capture and store the communication link signal between the drone and its remote controller, and then detect the drone target by detecting this link signal. Due to its advantages such as high detection sensitivity, low cost, relatively low susceptibility to external interference, and good concealment due to its passive nature, it has become one of the mainstream detection methods. Typical methods for RF signal-based drone detection include traditional detection methods and artificial intelligence-based detection methods.

[0003] Traditional detection methods, such as frequency domain analysis, time-spectrum analysis, and wavelet transform analysis, determine the presence of UAV targets by processing spectral data in different transform domains according to certain threshold criteria, including peak searching and signal parameter measurement and analysis. These methods employ a one-dimensional detection model based on noise floor statistics. Their performance depends on the detection threshold's ability to characterize and adapt to noise. In low signal-to-noise ratio environments, especially complex electromagnetic environments, their detection accuracy is low and their false alarm rate is high, failing to meet the application requirements for long-range UAV warning and countermeasures.

[0004] Artificial intelligence-based detection methods encompass two main approaches: machine learning and deep learning. Machine learning methods first extract various features from the signal using techniques such as Fourier transform, wavelet transform, or Hilbert transform, then train machine learning algorithms like random forests and SVMs to distinguish between drone signals and noise signals. Deep learning methods use radio frequency signals or transformed spectrum / time spectrum signals as data for network training and inference. These data-driven supervised learning methods, trained with a large number of drone signal data samples, can significantly improve the detection capability of known targets. However, their performance is entirely dependent on the diversity and comprehensiveness of the training samples, resulting in high sample data acquisition costs, high parameter tuning and training complexity, and the need for expensive hardware computing resources, making them unsuitable for detecting unknown targets.

[0005] Therefore, there is an urgent need to conduct research on high-probability detection technology for UAV targets in complex electromagnetic environments, to solve the problem of high false alarm rates of traditional detection algorithms in complex electromagnetic environments, and to improve detection accuracy while ensuring implementation cost and complexity. Summary of the Invention

[0006] To address the issues of low detection accuracy in complex electromagnetic environments or low signal-to-noise ratio conditions by traditional methods, as well as high training costs and poor generalization ability of supervised models, this invention provides an unsupervised clustering detection method and system for UAVs based on radio frequency signals. By using information entropy-based statistical denoising and multidimensional clustering processing, the detection accuracy in complex electromagnetic environments is improved while reducing implementation costs and computational load.

[0007] The first aspect of this invention proposes an unsupervised clustering detection method for unmanned aerial vehicles (UAVs) based on radio frequency signals, comprising: The radio frequency signals of the UAV were collected, and the time-frequency spectrum and time-frequency azimuth map of the UAV signal were obtained by sampling. The time-frequency azimuth matrix is ​​formed by merging the time-frequency spectrum diagram and the time-frequency azimuth diagram of the UAV signal. Denoising of the time-frequency azimuth matrix is ​​performed using a two-dimensional binning method; The denoised time-frequency azimuth matrix is ​​standardized, and several points are sampled to train a clustering model. The trained clustering model is then used to classify each point in the denoised time-frequency azimuth matrix. Based on the classification results, a time-azimuth-intensity map is generated from the frequency points contained in each category; A sliding window approach is used to denoise each time-azimuth-intensity map. Based on the motion attributes of the UAV target, static non-UAV targets are eliminated, and the target azimuth information is extracted to complete the target detection.

[0008] As a preferred embodiment, the acquisition of UAV radio frequency signals and the sampling to obtain the UAV signal time-frequency spectrum and the UAV signal time-frequency azimuth map specifically include: Radio frequency signals are continuously acquired for a certain duration, and initial time-frequency spectrum and initial time-frequency azimuth map are generated through short-time Fourier transform and real-time point-by-point direction finding module. Set a sampling rate c, and extract one row of data from the initial time spectrum map and the initial time-frequency azimuth map every c rows to finally form the UAV signal time spectrum map and the UAV signal time-frequency azimuth map.

[0009] As a preferred embodiment, the merging of the UAV signal time-frequency spectrum and the UAV signal time-frequency azimuth map to form a time-frequency azimuth matrix specifically includes: Each frequency point includes three attribute values: frequency value, signal strength value, and azimuth value. The three attribute values ​​are combined to form a three-dimensional matrix, and then expanded in the first and second dimensions to obtain the time-frequency azimuth matrix.

[0010] As a preferred embodiment, the denoising of the time-frequency azimuth matrix using a two-dimensional binning method specifically includes: For the time-frequency azimuth matrix, the intensity interval and frequency interval are divided into G small intervals at equal intervals, which are combined to form G×G intervals. The distribution of azimuth values ​​in the G×G intervals is then statistically analyzed. Calculate the entropy of the distribution in G×G intervals. If the entropy of the distribution is greater than the first preset threshold, delete the frequency points in the corresponding interval to complete the noise reduction.

[0011] As a preferred embodiment, the standardization process for the denoised time-frequency azimuth matrix specifically includes: The values ​​of frequency, intensity, and azimuth dimension in the denoised time-frequency azimuth matrix are standardized to normalize their range to between 0 and 1.

[0012] As a preferred embodiment, the resampling of several points trains a clustering model, and the trained clustering model is used to classify each point in the denoised time-frequency orientation matrix, specifically including: Set a maximum number of sampling points Q. For the standardized time-frequency azimuth matrix, its dimension is [W, 3], where W is the number of frequency points contained in a certain dimension. If W is greater than Q, then randomly select Q points from it to train the clustering algorithm and obtain the sampled matrix with a dimension of [Q, 3]. Set the range of the number of clusters for clustering [ , The algorithm iterates through the number of clusters within a set range and trains on the sampled matrix. - +1 clustering algorithm, and use silhouette coefficients in the standardized time-frequency orientation matrix to evaluate the clustering effect, and select the model with the highest silhouette coefficient as the final clustering model; The final clustering model is used to assign a category to each frequency point in the denoised time-frequency orientation matrix.

[0013] As a preferred embodiment, the step of generating a time-azimuth-intensity map from the frequency points contained in each category based on the classification results includes: Each frequency point in the time-frequency azimuth matrix is ​​distinguished according to different categories, and time-azimuth-intensity maps are drawn for each category.

[0014] As a preferred embodiment, the method of denoising each time-azimuth-intensity map using a sliding window approach specifically includes: For any given time-azimuth-intensity map, set a sliding window and slide it non-overlappingly on the time-azimuth-intensity map. Each slide cuts out a sub-time-azimuth-intensity map. Calculate the distribution of the maximum intensity value in the time dimension for each sub-time-azimuth-intensity map, and calculate the corresponding distribution entropy; If the distribution entropy corresponding to the sub-time-azimuth-intensity map is less than the second preset threshold, then the intensity value of the sub-time-azimuth-intensity map is set to zero; if there are no frequency points with an intensity value greater than 0 in the sub-time-azimuth-intensity map, then the sub-time-azimuth-intensity map is deleted. The above processing is performed on all time-azimuth-intensity maps to obtain denoised time-azimuth-intensity maps. Each time-azimuth-intensity map represents the trend of the target's azimuth and intensity changing over time.

[0015] The second aspect of this invention proposes an unsupervised clustering detection system for unmanned aerial vehicles based on radio frequency signals, comprising a computer system and a computer program, wherein the computer program executes the method proposed in the first aspect when the computer system is running.

[0016] Compared with existing technologies, the beneficial effects of adopting the above technical solution are as follows: The unsupervised radio frequency signal detection method proposed in this invention does not require offline model training. Through entropy-based automatic denoising and real-time multidimensional clustering, it significantly improves the target detection capability of UAVs in complex electromagnetic environments without increasing processing costs or increasing complexity. Attached Figure Description

[0017] Figure 1 This is a flowchart of the unsupervised clustering detection method for unmanned aerial vehicles based on radio frequency signals proposed in this invention.

[0018] Figure 2 This is a schematic diagram of a deployment implementation scheme in one embodiment.

[0019] Figure 3 This is a time-frequency spectrum diagram and a time-frequency orientation diagram in one embodiment.

[0020] Figure 4 This is a time-azimuth-intensity map converted from a sampled time-spectrum map and a time-frequency azimuth map in one embodiment.

[0021] Figure 5 This is a 3D visualization result of the time-frequency orientation matrix before and after denoising in one embodiment.

[0022] Figure 6 In one embodiment, after clustering, the points of each category are converted into a time-frequency-azimuth-intensity map.

[0023] Figure 7 This is the result of denoising the time-frequency-azimuth-intensity map for each category in one embodiment. Detailed Implementation

[0024] The embodiments of this application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar modules or modules having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. Rather, the embodiments of this application include all variations, modifications, and equivalents falling within the spirit and scope of the appended claims.

[0025] To address the issues of low detection accuracy in complex electromagnetic environments or low signal-to-noise ratio conditions using traditional methods, and the high training cost and poor generalization ability of supervised models, this invention proposes an unsupervised clustering detection method for UAVs based on radio frequency signals. By employing information entropy-based statistical denoising and multidimensional clustering processing, it improves detection accuracy in complex electromagnetic environments while reducing implementation costs and computational load. Please refer to [reference needed]. Figure 1 The specific plan is as follows: Step 1: Collect the UAV radio frequency signal and obtain the UAV signal time spectrum diagram and UAV signal time frequency azimuth diagram.

[0026] In this embodiment, a multi-channel direction-finding receiver continuously acquires radio frequency signals for a certain duration, and then generates an initial time spectrum diagram A through a short-time Fourier transform (STFT) and a real-time frequency-by-frequency point-finding module. and initial time-frequency azimuth map B Their dimensions are all [T] [, M], where T M is the number of spectral frames and M is the frame length.

[0027] Furthermore, a sampling rate c is set, and every c rows, the initial spectrum A is used as the sampling rate. and initial time-frequency azimuth map B A row of data is extracted to form the required UAV signal time-frequency spectrum diagram A and UAV signal time-frequency azimuth diagram B, both with dimensions [T, M], where T is as shown in equation (1): (1) Step 2: Combine the UAV signal time spectrum diagram and the UAV signal time-frequency azimuth diagram to form a time-frequency azimuth matrix.

[0028] For the obtained UAV signal time-frequency spectrum map A and UAV signal time-frequency azimuth map B, each frequency point contains three attribute values: frequency, signal strength, and azimuth. The frequency value is represented by the horizontal axis in both maps, the signal strength value by the vertical axis in map A, and the azimuth value by the vertical axis in map B. Combining these three attribute values ​​forms a new three-dimensional matrix C with dimensions [T, M, 3]. Expanding this matrix in the first and second dimensions yields the time-frequency azimuth matrix Ct. The dimension is [T×M, 3].

[0029] Step 3: Denoise the time-frequency azimuth matrix using a two-dimensional binning method.

[0030] Since the azimuth of a drone target does not change rapidly in a short period of time, the target's azimuth values ​​tend to be concentrated in three-dimensional space. Frequency points with excessively large deviations can be considered noise and can be used for noise reduction. Specifically: Step 3.1: In this embodiment, the time-frequency azimuth matrix C is adjusted in both the intensity and frequency dimensions. Create two-dimensional bins.

[0031] Take a binning parameter G, and divide both the intensity range and the frequency range into G smaller intervals at equal intervals, which are respectively ( , , …, )and( , , …, ), which together form G×G small intervals [( , ),( , ),…, ( , ), ( , ),…, ( , ),…,( , The distribution of azimuth values ​​within these G×G intervals is statistically analyzed (intervals are divided into 20 bins): (2) Step 3.2: Calculate the entropy of these distributions: (3) in, It is each distribution vector The distribution value.

[0032] Step 3.3: Set the first entropy threshold α. If the distribution... If the frequency point is greater than the first entropy threshold α, it indicates that the frequency point is noise. The corresponding frequency points in the interval are then removed to complete the denoising process. The denoised time-frequency orientation matrix C is then obtained. The result is matrix D with dimensions [W, 3], where W is the number of remaining points after denoising.

[0033] Step 4: Standardize the denoised time-frequency azimuth matrix, sample several points to train a clustering model, and then classify each point in the denoised time-frequency azimuth matrix using the trained clustering model.

[0034] Before model training, matrix standardization is required. In this embodiment, the values ​​of frequency, intensity, and orientation dimensions in matrix D are standardized to normalize their range to between 0 and 1, resulting in a new matrix D. The dimensions are [W, 3], as detailed below: (4) , , D respectively Mid-frequency value, intensity value, azimuth value.

[0035] For drone targets, the three attribute values—frequency, signal strength, and azimuth—do not differ significantly per unit time, and these points will be distributed in clusters or groups in space. In this embodiment, a maximum number of sampling points Q is set for a matrix D of dimension [W, 3]. If W is greater than Q, then Q points are randomly selected to train the k-means clustering algorithm, and the sampled matrix is ​​D. The dimension is [Q, 3].

[0036] The range of the number of clusters in the clustering is set to [ , ], ≥2, ≤8. Iterate through the number of clusters within the specified range. For each type of cluster count, in D... Training - +1 k-means clustering algorithm, in D The silhouette coefficient was used to evaluate the clustering effect, and the model with the highest silhouette coefficient was selected as the final clustering model. The number of clusters in this result was [number missing]. Each point in matrix D will be assigned a category by the clustering model.

[0037] Step 5: Based on the classification results, generate a time-azimuth-intensity map from the frequency points contained in each category; After determining the categories of all frequency points, each frequency point in matrix D is distinguished according to its category, and time-azimuth-intensity maps are plotted for each category: (5) The dimensions are all [T, AN], where AN represents the azimuth angle.

[0038] Step 6: Use a sliding window method to denoise each time-azimuth-intensity map, remove static non-UAV targets based on the motion attributes of the UAV target, extract the target azimuth information, and complete the target detection.

[0039] The actual entropy of the time-azimuth-intensity map generated in step 5 still contains some noise. In this embodiment, a method for denoising the time-azimuth-intensity map is proposed, specifically: Step 6.1: For any time-azimuth-intensity map Set a sliding window with width w and height h, in The system performs non-overlapping sliding, and each slide cuts out a small sub-time-azimuth-intensity map. .

[0040] Step 6.2: Create a sub-time-azimuth-intensity map. The intensity value in the value is calculated to its maximum value in each time dimension, resulting in... Distribution of maximum values ​​on the time axis Calculate each entropy : (6) in, yes The specific distribution values ​​in the data.

[0041] Step 6.3: Set the second entropy threshold β. If If it is less than β, then the sub-time-azimuth-intensity map will be... If all intensity values ​​are set to 0 and all frequencies are considered noise, then the time-azimuth-intensity map will be affected. If there are no points with an intensity value greater than 0, then the sub-time-azimuth-intensity map will be... Delete it, and it will be considered that there is no target in it.

[0042] Step 6.4: Generate all time-azimuth-intensity maps , , Perform steps 6.1 to 6.3 on all of them to obtain all the time-azimuth-intensity maps after denoising. Their dimensions are all [T, AN]. Each time-azimuth-intensity map can represent the trend of the target's azimuth and intensity changing over time.

[0043] Step 6.5: Combine all time-azimuth-intensity maps with the motion attributes of the UAV target, eliminate static non-UAV targets, extract target azimuth information, and complete target detection.

[0044] This invention also proposes an unsupervised clustering detection system for unmanned aerial vehicles (UAVs) based on radio frequency signals, comprising a computer system and a computer program, wherein the computer program executes the aforementioned method when the computer system runs.

[0045] To verify the effectiveness of this invention, the algorithm was deployed on a standard PC (containing an 8-core Intel i7 processor). A four-channel interferometer receiver was used to acquire radio frequency signals, generating time-spectrum data containing azimuth values ​​at each frequency point every 1 second, and storing it in real-time as a file. The PC processor reads the spectrum data file via a network and performs detection processing. Specific implementation details are as follows... Figure 2 As shown.

[0046] The 2.4GHz frequency band, characterized by high electromagnetic complexity, was selected as the experimental frequency band. The UAV link was controlled to operate within this band, which also included several unknown civilian communication signals. The continuous acquisition cycle was set to 1 second. The time-spectrum and time-frequency azimuth diagrams output by the four-channel interferometer receiver are shown below. Figure 3 As shown, (a) is the time-frequency spectrum diagram and (b) is the time-frequency azimuth diagram.

[0047] With a sampling rate of 10, the extracted time-frequency spectrum and time-frequency azimuth map are obtained. These are then converted into a time-azimuth-intensity map, as shown below. Figure 4 As shown in the image, there is a slanted azimuth line in the middle representing the drone target, but it is blended into the noise and cannot be easily extracted.

[0048] The sampled time-frequency spectrum and time-frequency azimuth map are converted into a time-frequency azimuth matrix. 10,000 points are randomly selected for visualization, and the results are as follows: Figure 5 As shown in (a). With an entropy threshold of α=1.8, the azimuth distribution vector and entropy of each cell in the time-frequency azimuth matrix are calculated using two-dimensional binning, and then denoising is performed. The denoising result is shown in (a). Figure 5 As shown in (b).

[0049] The denoised time-frequency azimuth matrix was normalized, and 1536 sample points were randomly selected for online model training using the k-means++ algorithm. The final iterative model was then used to cluster all time-frequency azimuth points, resulting in the following clustering pattern: Figure 6 As shown, a total of 5 cluster sets are formed.

[0050] Setting the entropy threshold β=3.5, denoising is performed again on each cluster set, resulting in the denoised time-frequency-azimuth-intensity map of each cluster set as follows: Figure 7 As shown, the positions of sets 1 and 2 remain unchanged over time, and are likely stationary targets; the position of set 3 changes dynamically over time, and can be identified as a moving target. Based on the application scenario, it is highly likely to be identified as a drone target; sets 4 and 5 are noise, which will be cleared directly after noise removal.

[0051] This completed a UAV detection process for handling clock cycles, verifying the invention's accurate detection capability for UAV targets in complex electromagnetic environments. A typical civilian UAV was selected for flight testing, with the link signal operating frequency band chosen at 2.4GHz, an area with extremely complex electromagnetic environments. Specific electromagnetic spectrum details are as follows... Figure 3 As shown. Using this method to detect UAV target signals, the detection accuracy reaches 90% and the false alarm rate is less than 10%; at the same time, the processing is highly real-time, and based on a normal PC environment (8-core Intel i7 processor), the single detection time does not exceed 1 second (the single radio frequency acquisition time is 1 second).

[0052] In particular, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts.

[0053] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0054] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0055] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0056] In another aspect, this application also provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the unsupervised clustering detection method and system for unmanned aerial vehicles based on radio frequency signals described in the above embodiments.

[0057] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the unsupervised clustering detection method and system for unmanned aerial vehicles based on radio frequency signals described in the above embodiments.

[0058] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0059] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0060] For those skilled in the art, the specific meanings of the above terms in this invention can be understood according to the specific circumstances; the accompanying drawings in the embodiments are used to clearly and completely describe the technical solutions in the embodiments of this invention. Obviously, the described embodiments are some embodiments of this invention, but not all embodiments. Generally, the components of the embodiments of this invention described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0061] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for unsupervised clustering detection of unmanned aerial vehicles (UAVs) based on radio frequency signals, characterized in that, include: The radio frequency signals of the UAV were collected, and the time-frequency spectrum and time-frequency azimuth map of the UAV signal were obtained by sampling. The time-frequency azimuth matrix is ​​formed by merging the time-frequency spectrum diagram and the time-frequency azimuth diagram of the UAV signal. Denoising of the time-frequency azimuth matrix is ​​performed using a two-dimensional binning method; The denoised time-frequency azimuth matrix is ​​standardized, and several points are sampled to train a clustering model. The trained clustering model is then used to classify each point in the denoised time-frequency azimuth matrix. Based on the classification results, a time-azimuth-intensity map is generated from the frequency points contained in each category; A sliding window method is used to denoise each time-azimuth-intensity map. Based on the motion attributes of the UAV target, static non-UAV targets are removed, and the target azimuth information is extracted to complete the target detection. The denoising of the time-frequency azimuth matrix using a two-dimensional binning method specifically includes: For the time-frequency azimuth matrix, the intensity interval and frequency interval are divided into G smaller intervals at equal intervals, and then combined to form G×G intervals. The distribution of azimuth values ​​in the G×G intervals is then statistically analyzed. Calculate the entropy of the distribution in G×G intervals. If the entropy of the distribution is greater than the first preset threshold, delete the frequency points in the corresponding interval to complete the noise reduction. The method of denoising each time-azimuth-intensity map using a sliding window technique specifically includes: For any given time-azimuth-intensity map, set a sliding window and slide it non-overlappingly on the time-azimuth-intensity map. Each slide cuts out a sub-time-azimuth-intensity map. Calculate the distribution of the maximum intensity value in the time dimension for each sub-time-azimuth-intensity map, and calculate the corresponding distribution entropy; If the distribution entropy corresponding to the sub-time-azimuth-intensity map is less than the second preset threshold, then the intensity value of the sub-time-azimuth-intensity map is set to zero; if there are no frequency points with an intensity value greater than 0 in the sub-time-azimuth-intensity map, then the sub-time-azimuth-intensity map is deleted. The above processing is performed on all time-azimuth-intensity maps to obtain denoised time-azimuth-intensity maps. Each time-azimuth-intensity map represents the trend of the target's azimuth and intensity changing over time.

2. The unsupervised clustering detection method for UAVs based on radio frequency signals according to claim 1, characterized in that, The acquisition of UAV radio frequency signals, and the sampling to obtain the UAV signal time-frequency spectrum and UAV signal time-frequency azimuth map, specifically includes: Radio frequency signals are continuously acquired for a certain duration, and initial time-frequency spectrum and initial time-frequency azimuth map are generated through short-time Fourier transform and real-time point-by-point direction finding module. Set a sampling rate c, and extract one row of data from the initial time spectrum map and the initial time-frequency azimuth map every c rows to finally form the UAV signal time spectrum map and the UAV signal time-frequency azimuth map.

3. The unsupervised clustering detection method for UAVs based on radio frequency signals according to claim 1, characterized in that, The merged UAV signal time-frequency spectrum and UAV signal time-frequency azimuth map form a time-frequency azimuth matrix, specifically including: Each frequency point includes three attribute values: frequency value, signal strength value, and azimuth value. The three attribute values ​​are combined to form a three-dimensional matrix, and then expanded in the first and second dimensions to obtain the time-frequency azimuth matrix.

4. The unsupervised clustering detection method for UAVs based on radio frequency signals according to claim 1, characterized in that, The standardization process for the denoised time-frequency azimuth matrix specifically includes: The values ​​of frequency, intensity, and azimuth dimension in the denoised time-frequency azimuth matrix are standardized to normalize their range to between 0 and 1.

5. The unsupervised clustering detection method for unmanned aerial vehicles based on radio frequency signals according to claim 1, characterized in that, The process of resampling several points to train a clustering model, and then using the trained clustering model to classify each point in the denoised time-frequency orientation matrix, specifically includes: Set a maximum number of sampling points Q. For the standardized time-frequency azimuth matrix, its dimension is [W, 3], where W is the number of frequency points contained in a certain dimension. If W is greater than Q, then randomly select Q points from it to train the clustering algorithm and obtain the sampled matrix with a dimension of [Q, 3]. Set the range of the number of clusters for clustering [ , The algorithm iterates through the number of clusters within a set range and trains on the sampled matrix. - +1 clustering algorithm, and use silhouette coefficients in the standardized time-frequency orientation matrix to evaluate the clustering effect, and select the model with the highest silhouette coefficient as the final clustering model; The final clustering model is used to assign a category to each frequency point in the denoised time-frequency orientation matrix.

6. The unsupervised clustering detection method for unmanned aerial vehicles based on radio frequency signals according to claim 1, characterized in that, The step of generating a time-azimuth-intensity map from the frequency points included in each category based on the classification results includes: Each frequency point in the time-frequency azimuth matrix is ​​distinguished according to different categories, and time-azimuth-intensity maps are drawn for each category.

7. A drone unsupervised clustering detection system based on radio frequency signals, characterized in that, It includes a computer system and a computer program, wherein the computer program executes the method as described in any one of claims 1 to 6 when the computer system is running.

Images