Unmanned aerial vehicle signal intelligent identification method and system based on RFSOC, and storage medium

Abstract

The invention discloses an RFSOC-based unmanned aerial vehicle signal intelligent identification method and system, and a storage medium, and belongs to the technical field of unmanned aerial vehicle identification. The method comprises the following steps: a data acquisition stage: acquiring a plurality of unmanned aerial vehicle signals in a plurality of data sources to obtain an unmanned aerial vehicle data set of which the data volume is greater than a preset data volume; a preprocessing stage: preprocessing the unmanned aerial vehicle data set to obtain an unmanned aerial vehicle radio frequency fingerprint feature data set; in the model training stage, the unmanned aerial vehicle radio frequency fingerprint feature data set is input into a deep learning network for training and testing, and a trained unmanned aerial vehicle radio frequency fingerprint recognition model is obtained; and an identification stage: deploying an unmanned aerial vehicle radio frequency fingerprint identification model, and inputting the to-be-identified signal into the unmanned aerial vehicle radio frequency fingerprint identification model to obtain an identification result. According to the invention, multiple unmanned aerial vehicles can be detected and identified in real time, missed judgment of results caused by detection and identification of a single scene and a fixed frequency point is avoided, and the safety is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) identification, and particularly to an intelligent UAV signal identification method, system and storage medium based on RFSOC. Background Art

[0002] In recent years, with the rapid development of UAV technology, UAVs have been widely used in various military and civilian fields, including agricultural irrigation, aerial photography mapping, mine detection and other fields. While UAVs are widely used in various fields, they also pose great potential threats to public safety. Therefore, how to achieve the detection, identification and interference countermeasures of UAVs is an urgent technical problem to be solved.

[0003] Considering the advantages and disadvantages of various detection means comprehensively, the detection means based on the signal characteristics of the UAV image link and the remote control link is an ideal detection means in the urban environment. However, the traditional detection means based on signal characteristics has a long time-consuming and low recognition probability, and it is difficult to meet the needs of UAV signal identification and detection. The UAV signal detection based on artificial intelligence is the ideal means in the current UAV signal detection and identification field. By extracting the fingerprint features of the collected UAV signals and comparing the fingerprint features with the features in the database, the identification of various civilian UAVs with signals in the 2.4GHz or 5.8GHz frequency bands can be achieved. This method is difficult to identify racing drones or privately modified UAVs, which brings great potential threats to the specific applications in military and civilian fields. The currently proposed UAV radio frequency identification means has obvious limitations in actual complex scenarios, and is large in volume, expensive in price, and there is no method and device for large-bandwidth UAV identification that can run offline on RFSOC. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and provide an intelligent UAV signal identification method, system and storage medium based on RFSOC.

[0005] The purpose of the present invention is achieved by the following technical solutions: In the first aspect of the present invention, an intelligent UAV signal identification method based on RFSOC is provided, including the following steps: In the data acquisition stage, various UAV signals in multiple data sources are collected to obtain a UAV data set with a data volume greater than a preset data volume; In the preprocessing stage, the UAV data set is preprocessed to obtain a UAV radio frequency fingerprint feature data set; In the model training stage, the UAV radio frequency fingerprint feature data set is input into a deep learning network for training and testing to obtain a trained UAV radio frequency fingerprint recognition model; In the identification stage, the UAV radio frequency fingerprint recognition model is deployed, and the signal to be identified is input into the UAV radio frequency fingerprint recognition model to obtain an identification result.

[0006] Preferably, the UAV signals are video transmission signals, remote control signals, and environmental signals.

[0007] Preferably, the preprocessing stage further includes the following steps: Perform time-frequency transformation on the UAV signals to enhance the fingerprint features of the signals, obtaining a UAV radio frequency fingerprint feature dataset. Let the I / Q data of the UAV dataset at the time point n be x ( n ). For the UAV dataset at N time points, the preprocessing method is as follows: ; where X i is the result of the short-time Fourier transform of the UAV dataset; is the Hamming window function; R represents the number of overlapping samples between adjacent windows; w represents the window function; m is the index value of the window function; k is the index value of the frequency, k = 1, 2... N .

[0008] Preferably, the m is 256 or 128.

[0009] Preferably, the model training stage further includes the following steps: Classify the UAV signals and noise signals according to a first preset ratio to obtain a training sample set and a test sample set; Perform STFT transformation on the samples in the training sample set, and divide the data in the training sample set into a training set and a validation set according to a second preset ratio; Define labels for the signals in the training set, classifying them as UAVs and non-UAVs; Train and test the deep learning network according to the validation set, test sample set, and the training set with defined labels to obtain a trained UAV radio frequency fingerprint recognition model.

[0010] Preferably, for the deployment of the UAV radio frequency fingerprint recognition model, the following steps are further included: Convert the UAV radio frequency fingerprint recognition model into an MNN model, and then deploy it to the ARM side of the RFSOC chip. Receive real-time data transmitted by the AXI-DMA bus of the RFSOC chip through the ARM side to complete the lightweight deployment of the UAV radio frequency fingerprint recognition model.

[0011] Preferably, the recognition stage further includes the following steps: Receive the signal to be recognized within the preset frequency range and transmit it to the digital processing board of the RFSOC chip through a radio frequency cable; Complete signal conditioning through the conditioning unit, and then perform AD sampling at a preset sampling rate to obtain I / Q data; based on a preset bandwidth, perform digital down-conversion processing on multiple signals; the digital down-conversion processing is to program inside the RFSOC chip to transform the I / Q data into baseband I / Q signals through mixing, decimation, and filtering, and perform signal detection on the baseband I / Q signals to confirm the existence of the signal and then cache it into the memory; After performing STFT transformation on the baseband I / Q signals, take multiple inputs simultaneously and input them into the UAV radio frequency fingerprint recognition model for fingerprint feature matching. Continuously recognize the same signal multiple times. If the number of times recognized as a UAV is greater than the preset number of times, then consider this signal as a UAV signal; Finally, obtain the recognition result and display it.

[0012] Preferably, the deep learning network is ResNet50.

[0013] The second aspect of the present invention provides: A UAV signal intelligent recognition system based on RFSOC, used to implement any of the above UAV signal intelligent recognition methods based on RFSOC, including: an antenna, the antenna is connected to a full-band data receiving device, and the full-band data receiving device is connected to a display device and a memory through a high-speed data interface; the antenna is used to receive signals within the preset frequency range; the full-band data receiving device uses an RFSOC chip to complete signal sampling, and the sampled signal is sent to the ARM end of the RFSOC chip through down-conversion and preprocessing through a high-speed bus. The ARM end calls the UAV radio frequency fingerprint recognition model to perform matching and calculation processing of UAV radio frequency fingerprint features, and finally outputs the recognition result of whether the signal is a UAV signal.

[0014] The third aspect of the present invention provides: A computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, any of the above UAV signal intelligent recognition methods based on RFSOC is implemented.

[0015] The beneficial effects of the present invention are: 1) Completed the offline RFSOC deployment of the UAV radio frequency fingerprint recognition model, and the RFSOC chip integrates a high-speed AD inside, which can effectively reduce the research and development cost and reduce the volume of the equipment.

[0016] 2) It can not only detect and identify various civilian fixed-frequency drones in real time, but also detect and identify signals of other assembled drones, racing drones or military drones in the range of 300 MHz to 6 GHz, avoiding missed judgments in the detection and identification of single scenarios and fixed frequencies, with higher security.

[0017] 3) The RFSOC chip is adopted to achieve high-speed sampling of wide-bandwidth RF signals, which can effectively reduce the frequency conversion link, high-speed AD circuit and matching link, reduce the volume of the device and lower the cost of the device. In addition, in the implementation process of this application, the UAV RF fingerprint recognition model is deployed on the RFSOC chip, which can realize the transmission of baseband data in the full frequency band (300 MHz to 6 GHz), and at the same time can reduce problems such as high-speed data transmission and model deployment brought by the current computer integrated GPU and other calculations. Description of the Drawings

[0018] Figure 1 is a flowchart of an intelligent UAV signal recognition method based on RFSOC; Figure 2 is a schematic block diagram of an intelligent UAV signal recognition system based on RFSOC; In the figure, 1001 is an antenna; 1002 is a full-band data receiving device; 1003 is a high-speed data interface; 1004 is a memory; 1005 is a display device. Detailed Embodiment

[0019] Next, the technical solutions of the present invention will be described clearly and completely in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0020] Refer to Figure 1 - Figure 2 , the first aspect of the present invention provides: an intelligent UAV signal recognition method based on RFSOC, including the following steps: In the data acquisition stage, various UAV signals in multiple data sources are collected to obtain a UAV data set with a data volume greater than the preset data volume; In the preprocessing stage, the UAV data set is preprocessed to obtain a UAV RF fingerprint feature data set; In the model training stage, the UAV RF fingerprint feature data set is input into a deep learning network for training and testing to obtain a trained UAV RF fingerprint recognition model; In the recognition stage, the UAV RF fingerprint recognition model is deployed, and the signal to be recognized is input into the UAV RF fingerprint recognition model to obtain a recognition result.

[0021] In this embodiment, the data source includes UAV data sources for various commonly used signals at home and abroad, and it is necessary to collect more than 1000 complete sample data of UAV signals. In some embodiments, the UAV signals are video transmission signals, remote control signals, and environmental signals.

[0022] In some embodiments, the preprocessing stage further includes the following steps: Perform time-frequency transformation on the UAV signal to enhance the fingerprint feature of the signal to obtain a UAV radio frequency fingerprint feature dataset. Let the I / Q data of the UAV dataset be at the time point n The value at is x ( n ). For the UAV dataset at N time points, the preprocessing method is as follows: ; Among them, X i Is the result after short-time Fourier transform of the UAV dataset; Is the Hamming window function; R Represents the number of overlapping samples between adjacent windows; w Represents the window function; m Is the index value of the window function; k Is the index value of the frequency, k = 1, 2... N .

[0023] In this embodiment, m The value of is generally 2 n .

[0024] In some embodiments, the m Is 256 or 128.

[0025] In this embodiment, in order to ensure real-time processing and accuracy during the implementation process m Select 256 or 128.

[0026] In some embodiments, the model training stage further includes the following steps: Classify the UAV signal and the noise signal according to the first preset ratio to obtain a training sample set and a test sample set; Perform STFT transformation on the samples in the training sample set, and divide the data in the training sample set into a training set and a validation set according to the second preset ratio; Define labels for the signals in the training set, and classify them into UAVs and non-UAVs; Train and test the deep learning network according to the validation set, the test sample set, and the training set with defined labels to obtain a trained UAV radio frequency fingerprint recognition model.

[0027] In this embodiment, the deep learning network is ResNet50, which is a deep convolutional neural network based on the residual network architecture. Its core feature is that it can meet the deployment requirements of miniaturization, lightweight, and real-time processing. It includes an input layer: data of 256x256; a convolutional layer: to complete the preliminary extraction of radio frequency fingerprint features, with a stride of 3; residual modules: 32 residual modules; a global average pooling layer: after all the residual modules, convert the feature map into a vector; a fully connected layer: finally, a fully connected layer is used to output the classification result.

[0028] In some embodiments, the steps for deploying the UAV radio frequency fingerprint recognition model further include the following: Convert the UAV radio frequency fingerprint recognition model to an MNN model, and then deploy it to the ARM side of the RFSOC chip. The ARM side receives real-time data transmitted through the AXI-DMA bus of the RFSOC chip to complete the lightweight deployment of the UAV radio frequency fingerprint recognition model.

[0029] In this embodiment, the RFSOC chip is used to complete the miniaturization, lightweight deployment, and real-time processing of the model, without the need for large computing power and storage space. The RFSOC chip includes, but is not limited to, RFSOC chips from the United States or Fudan Microelectronics Company, including, but not limited to, RFSOC47DR, RFSOC48DR, FMZQ28DR, etc. It has a high-speed ADC and integrates an RFSCO chip with an ARM core. The PL side of the RFSOC chip can receive ADC data, and the PS side contains an ARM core that can complete the deployment of the model. The PL side and the PS side have high-speed interfaces and can achieve high-speed transmission of IQ data.

[0030] In some embodiments, the recognition stage further includes the following steps: Receive the signal to be recognized within the preset frequency range and transmit it to the digital processing board of the RFSOC chip through a radio frequency cable; Complete signal conditioning through a conditioning unit, and then perform AD sampling at a preset sampling rate to obtain I / Q data; based on a preset bandwidth, perform digital down-conversion processing on multiple signals; the digital down-conversion processing is to program inside the RFSOC chip to transform the I / Q data into a baseband I / Q signal through mixing, decimation, and filtering, and perform signal detection on the baseband I / Q signal to confirm the existence of the signal and then cache it in memory; After performing STFT transformation on the baseband I / Q signal, take multiple inputs simultaneously and input them into the UAV radio frequency fingerprint recognition model for fingerprint feature matching. Continuously recognize the same signal multiple times. If the number of times recognized as a UAV is greater than the preset number of times, then the signal is considered a UAV signal; Finally, the recognition result is obtained and displayed.

[0031] In this embodiment, the preset frequency range is 300 MHz to 6 GHz. The preset sampling rate is 5 GSPS, which can instantaneously complete the sampling of signals with a maximum bandwidth of 5 GHz. The preset bandwidth is 80 MHz, and digital down-conversion processing of 57 signals is performed based on the 80 MHz bandwidth. Signal detection is performed to reduce the amount of data cached in the DDR4 memory. In order to reduce the misrecognition of UAV fingerprint signals and reduce the false alarm rate of UAV signals, the same signal is continuously recognized 5 times here. If the signal is recognized as a UAV 3 times, it is determined that the baseband I / Q signal is a UAV signal and the recognition result is output; otherwise, no result is output. To ensure the processing speed of broadband signals, the UAV radio frequency fingerprint recognition model has multiple channels and can simultaneously input multiple baseband I / Q signals after STFT transformation. The recognition result of the UAV is sent to the local computer through the network port built in the RFSOC chip for displaying the recognition result.

[0032] In some embodiments, the deep learning network is ResNet50.

[0033] The second aspect of the present invention provides: A UAV signal intelligent recognition system based on RFSOC for implementing any one of the above UAV signal intelligent recognition methods based on RFSOC, including: an antenna 1001, the antenna 1001 is connected to a full-band data receiving device 1002, and the full-band data receiving device 1002 is connected to a display device 1005 and a memory 1004 through a high-speed data interface 1003; the antenna 1001 is used to receive signals within a preset frequency range; the full-band data receiving device 1002 uses an RFSOC chip to complete signal sampling, and the sampled signal is sent to the ARM end of the RFSOC chip through a high-speed bus after down-conversion and preprocessing. The ARM end calls the UAV radio frequency fingerprint recognition model to perform matching and calculation processing of UAV radio frequency fingerprint features, and finally outputs the recognition result of whether the signal is a UAV signal.

[0034] In this embodiment, the specific selections for each part are as follows: The antenna 1001 selects a broadband omnidirectional antenna with a frequency range of 300 MHz to 6 GHz, and the polarization mode is vertical polarization; The full-band data receiving device 1002 selects the XCZU47DR of the RFSOC chip. This chip integrates 8 high-speed ADCs, with a maximum sampling rate of 5 GSPS, and can realize the sampling of signals from 9 kHz to 6 GHz. The RFSOC chip also includes rich programmable logic resources and 4 ARM A56 processing cores. In terms of interfaces, DP interfaces, gigabit Ethernet interfaces, IO interfaces, etc. are reserved for the transmission of data results and the expansion of functions. The memory 1004 has been integrated into the RFSOC chip of the full-band data receiving device 1002 in this embodiment. In this embodiment, the display device 1005 selects a computer terminal with a gigabit network port. The radio frequency signal received by the 300 MHz to 6 GHz broadband omnidirectional antenna directly enters the high-speed RFSOC chip. The high-speed AD integrated inside the RFSOC chip completes the real-time sampling of high-sampling and large-bandwidth signals. After digital down-conversion, the sampling rate is reduced and sent to the integrated ARM core unit of the RFSOC chip through the high-speed bus AXI-DMA interface for radio frequency fingerprint identification processing of the UAV signal. Finally, the result is sent to the display device 1005 for result display. The processing methods of the ARM core unit include running bare metal, Linux, etc. In this embodiment, the Linux operating system is used to perform relevant file management and data management methods for processing.

[0035] The third aspect of the present invention provides: A computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, any of the above-mentioned UAV signal intelligent recognition methods based on RFSOC is realized.

[0036] The above are only the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be changed within the scope of the concept described herein through the above teachings or the technology or knowledge in related fields. And the changes and alterations made by those skilled in the art without departing from the spirit and scope of the present invention should all be within the protection scope of the appended claims of the present invention.

Claims

1. An intelligent recognition method for drone signals based on RFSOC, characterized in that: It includes the following steps: In the data acquisition stage, various UAV signals in multiple data sources are collected to obtain a UAV dataset with a data volume greater than a preset data volume; In the preprocessing stage, the UAV dataset is preprocessed to obtain a UAV radio frequency fingerprint feature dataset; In the model training stage, the UAV radio frequency fingerprint feature dataset is input into a deep learning network for training and testing to obtain a trained UAV radio frequency fingerprint recognition model; In the recognition stage, the UAV radio frequency fingerprint recognition model is deployed, and the signal to be recognized is input into the UAV radio frequency fingerprint recognition model to obtain a recognition result.

2. The intelligent recognition method for UAV signals based on RFSOC according to claim 1, characterized in that: The UAV signals are video transmission signals, remote control signals, and environmental signals.

3. The intelligent identification method of UAV signals based on RFSOC according to claim 1, characterized in that: The preprocessing stage further includes the following steps: Perform time-frequency transformation on the UAV signal to enhance the fingerprint features of the signal, obtaining the UAV radio frequency fingerprint feature dataset. Let the I / Q data of the UAV dataset take the value at the time point n be x ( n ). For the UAV dataset at N time points, the preprocessing method is as follows: ； Among them, X i is the result after short-time Fourier transform of the UAV dataset; is the Hamming window function; R represents the number of overlapping samples between adjacent windows; w represents the window function; m is the index value of the window function; k is the index value of the frequency, k = 1, 2... N .

4. The method for intelligent recognition of UAV signals based on RFSOC according to claim 3, characterized in that: The described m is 256 or 128.

5. The intelligent recognition method of drone signals based on RFSOC according to claim 1, wherein: The model training stage further includes the following steps: Classify the UAV signals and noise signals according to a first preset ratio to obtain a training sample set and a test sample set; Perform STFT transformation on the samples in the training sample set, and divide the data in the training sample set into a training set and a validation set according to a second preset ratio; Define labels for the signals in the training set, which are divided into UAVs and non-UAVs; Train and test the deep learning network according to the validation set, the test sample set, and the training set with defined labels to obtain a trained UAV radio frequency fingerprint recognition model.

6. The intelligent identification method for UAV signals based on RFSOC according to claim 1, wherein: The deployment of the UAV radio frequency fingerprint recognition model further includes the following steps: Convert the UAV radio frequency fingerprint recognition model into an MNN model, and then deploy it to the ARM side of the RFSOC chip. The ARM side receives real-time data transmitted by the AXI-DMA bus of the RFSOC chip to complete the lightweight deployment of the UAV radio frequency fingerprint recognition model.

7. The intelligent recognition method for UAV signals based on RFSOC according to claim 1, characterized in that: The recognition stage further includes the following steps: Receive the signal to be recognized within a preset frequency range and transmit it to the digital processing board of the RFSOC chip through a radio frequency cable; Complete signal conditioning through a conditioning unit, and then perform AD sampling at a preset sampling rate to obtain I / Q data; based on a preset bandwidth, perform digital down-conversion processing on multiple signals; the digital down-conversion processing is to program inside the RFSOC chip to transform the I / Q data into a baseband I / Q signal through mixing, decimation, and filtering, and cache the baseband I / Q signal into memory after signal detection to confirm the existence of the signal; After performing STFT transformation on the baseband I / Q signal, input it in multiple channels simultaneously, input it into the UAV radio frequency fingerprint recognition model for fingerprint feature matching, and continuously recognize the same signal multiple times. If the number of times recognized as a UAV is greater than a preset number, then the signal is considered a UAV signal; Finally, obtain the recognition result and display it.

8. The intelligent recognition method for UAV signals based on RFSOC according to any one of claims 1-7, characterized in that: The deep learning network is ResNet50.

9. An intelligent UAV signal recognition system based on RFSOC, characterized in that: A method for realizing intelligent identification of UAV signals based on RFSOC as described in any one of claims 1-8, comprising: an antenna, the antenna is connected to a full-band data receiving device, and the full-band data receiving device is connected to a display device and a memory through a high-speed data interface; the antenna is used to receive signals within a preset frequency range; the full-band data receiving device uses an RFSOC chip to complete signal sampling, and the sampled signal is sent to the ARM end of the RFSOC chip through down-conversion and preprocessing through a high-speed bus. The ARM end calls the UAV radio frequency fingerprint recognition model to perform matching and calculation processing on the UAV radio frequency fingerprint features, and finally outputs the recognition result of whether the signal is a UAV signal.

10. A computer-readable storage medium, characterized in that: Computer-executable instructions are stored in the computer-readable storage medium, and when the computer-executable instructions are loaded and executed by a processor, the method for realizing intelligent identification of UAV signals based on RFSOC as described in any one of claims 1-8 is realized.

Images