Unmanned aerial vehicle radio frequency signal identification method and system based on multi-station fusion signal
By deploying the receiving stations in multiple geographical locations and using LSTMs optimized by timing attention mechanisms to perform drone radio frequency signal processing, combined with the signal-to-noise ratio of the signal-to-noise ratio of the signal-to-noise ratio of the signal-to-noise ratio in the signal-to-noise is imposed, the problem of difficulty in achieving accurate identification in complex environments in the prior art is solved, and the accuracy and robustness of the identification are improved.
Patent Information
- Application Number
- CN202411602337.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-11
- Publication Date
- 2025-02-28
AI Technical Summary
Existing drone radio frequency signal recognition technology is difficult to achieve accurate identification in complex and changeable environments, especially in multi-signal, multi-path and high-interference environments. The dependence of a single signal source limits the performance and accuracy of the identification system.
By deploying the receiving stations in multiple geographical locations, collecting drone radio frequency signals and their signal-to-noise ratios in different directions and distances, using LSTM optimized by timing attention mechanism for signal processing and identification, and combining the signal-to-noise ratio of the signal-to-noise ratio of the signal-to-noise ratio of the signal-to-noise ratio of the signal-to-noise is given to determine the final identification result.
It improves the accuracy and robustness of the recognition of radio frequency signals of the drone, enhances the system's spatial coverage ability of the signal source, optimizes resource utilization, and ensures the dominant role of high-quality signals in the recognition process.
Smart Images
Figure CN119538190A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of signal recognition technology, and in particular to a method and system for identifying unmanned aerial vehicle radio frequency signals based on multi-station fusion signals. Background Art
[0002] With the rapid development of technology, drones have been widely used in many fields such as military reconnaissance, terrain mapping, and agricultural monitoring. However, with the popularization of drone use, the safe management and monitoring of drones has become a major challenge. Especially in dense urban environments, illegal or hostile drones may pose a threat to people's property safety and privacy. Therefore, an effective drone radio frequency (RF) signal recognition system is crucial to ensure the safe flight and mission execution of drones. At present, the recognition of drone RF signals mainly relies on traditional signal processing technology and pattern recognition methods. These methods often rely on pre-defined feature extraction and static recognition algorithms, which are difficult to adapt to complex and changing environmental conditions and signal interference. In addition, the reliance on a single signal source limits the performance and accuracy of the recognition system, especially in multi-signal, multi-path and high-interference environments.
[0003] To overcome these limitations, researchers have begun to explore the use of artificial intelligence technology, especially recurrent neural networks (RNNs), to improve the recognition of drone RF signals. The advantage of RNNs is that they can process sequence data and learn the temporal dynamic characteristics of signals, thereby achieving more accurate signal recognition in complex environments. However, even RNN-based methods face challenges in data fusion and processing large-scale signal inputs. Summary of the invention
[0004] The embodiment of the present application provides a method and system for identifying UAV radio frequency signals based on multi-station fusion signals. By aggregating signal inputs from multiple stations, the diversity of data and the robustness of signals are increased. The LSTM optimized by the temporal attention mechanism can deeply analyze and process the time series features in the signal, greatly improving the accuracy of recognition.
[0005] The present application embodiment provides a method for identifying UAV radio frequency signals based on multi-station fusion signals, including:
[0006] Deploy receiving stations at multiple geographical locations to capture drone RF signals and their signal-to-noise ratio (SNR) from different directions and distances;
[0007] Perform signal processing based on the acquired drone RF signal and its SNR, and align the RF signal;
[0008] Inputting the radio frequency signal of each station after signal processing into a recurrent neural network LSTM to utilize the LSTM to output the category to which the signal belongs, wherein the LSTM is combined with a temporal attention mechanism;
[0009] The signal category of each station output by LSTM is weighted according to the corresponding SNR to determine the final signal recognition result, where the signal with a higher weight accounts for a greater proportion in the fusion result.
[0010] Optionally, signal processing based on the acquired drone RF signal and its SNR includes:
[0011] The acquired UAV radio frequency signal is amplified using an automatic gain control (AGC) amplifier;
[0012] The amplified radio frequency signal is low-pass filtered using a low-pass filter;
[0013] Performing denoising on the low-pass filtered RF signal;
[0014] The denoised RF signals are aligned in time using a cross-correlation function.
[0015] Optionally, the RF signals of each station after signal processing are input into a recurrent neural network LSTM, so as to utilize the categories to which the LSTM output signals belong, including:
[0016] Arrange the IQ sequence of the RF signal after signal processing into a set format (B, 1, H*W), where B represents the batch size, 1 represents the number of channels of the signal sequence, H*W represents the length of the sequence, and W is the width of the two-dimensional data;
[0017] The sorted IQ sequence in a set format is input into the LSTM, wherein the temporal attention mechanism of the LSTM is implemented by the attention weight, and the attention weight α t The attention score function is used to calculate and satisfy:
[0018]
[0019] Among them, s t-1 is the hidden state of LSTM at time point t-1, h t is the input at time point t, score is the function used to calculate the attention score;
[0020] After being processed by the LSTM, the output of each time point is weighted and summed according to its attention weight to form a final representation, and the final representation is classified through a fully connected layer.
[0021] Optionally, the signal category of each station output by LSTM is weighted according to the corresponding SNR to determine the final signal recognition result including:
[0022] Calculate the weight W of each decision based on the SNR value of each site i , to normalize the SNR value:
[0023]
[0024] The signal recognition result is determined by weighted averaging based on the calculated weights of the signal sources at each site.
[0025] Optionally, determining the signal identification result by weighted average according to the calculated weights of the signal sources of each site includes:
[0026] The decision of each station is defined as the probability P of the signal belonging to a certain category. i (c), where c represents the possible category, then the weighted average category probability satisfies:
[0027]
[0028] According to the weighted average category probability, the final recognition category is determined based on the entropy weight adjustment mechanism.
[0029] Optionally, according to the weighted average category probability, the final recognition category is determined based on the entropy weight adjustment mechanism, including:
[0030] Using the entropy value H i To measure the uncertainty of each receiving station’s decision, satisfying:
[0031]
[0032] According to the entropy value H i Adjust the weight of each receiving station, the adjusted weight w i ′ for:
[0033]
[0034] By selecting the category with the highest weighted average probability as the final recognition category
[0035] c=arg max c P(c).
[0036] Optionally, deploying the receiving stations at multiple geographical locations includes arranging the receiving stations to cover possible activity ranges and flight paths of the drone.
[0037] The embodiment of the present application also proposes a UAV radio frequency signal recognition system based on multi-station fusion signals, which is characterized by comprising a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the UAV radio frequency signal recognition method based on multi-station fusion signals are implemented as described above.
[0038] The embodiment of the present application increases the diversity of data and the robustness of signals by aggregating signal inputs from multiple sites, and the LSTM optimized through the temporal attention mechanism can deeply analyze and process the time series features in the signal, greatly improving the recognition accuracy.
[0039] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:
[0041] Figure 1 The following is a flow chart of a method for identifying UAV radio frequency signals based on multi-station fusion signals according to an embodiment of the present application;
[0042] Figure 2 The figure is a schematic diagram of the identification network structure of an embodiment of the present application. DETAILED DESCRIPTION
[0043] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0044] The present application embodiment provides a method for identifying UAV radio frequency signals based on multi-station fusion signals, such as Figure 1 As shown, including:
[0045] In step S101, receiving stations are deployed at multiple geographical locations to capture the RF signals of drones in different directions and at different distances and their signal-to-noise ratios (SNR). In some embodiments, deploying receiving stations at multiple geographical locations includes arranging receiving stations to cover the possible activity range and flight path of the drone, thereby ensuring that no matter where the drone is, the signal it sends can be captured by at least one receiving station. Such a multi-station layout not only reduces blind spots, but also ensures continuous monitoring of drone activities. The receiving station is usually equipped with a highly sensitive receiving device that can process weak RF signals and has the ability to distinguish signals from different drones, and can maintain a high recognition accuracy even in a complex electromagnetic environment. The signal-to-noise ratio (SNR), as an important indicator for measuring the quality of the received signal, plays a key role in this process. A high SNR means good signal quality, the amount of information is greater than the noise level, and it is more conducive to subsequent signal processing and recognition. Therefore, when designing the receiving station, it is necessary not only to consider the signal capture capability, but also to be able to effectively measure and optimize the SNR, and to adjust the receiving parameters to adapt to changes in environmental noise and signal strength to ensure the best reception quality.
[0046] Through this signal acquisition method, the system can obtain multi-angle views of the same signal source from multiple directions, increasing the diversity of the signal. This is particularly important for dealing with signal deformation caused by problems such as occlusion and multipath propagation, because the fusion of multi-source signal data can improve the accuracy and reliability of recognition. In addition, the multi-site layout also optimizes the comprehensive monitoring of drone activities in the target area, greatly improving the system's spatial coverage of the signal source, thereby improving the robustness of the signal.
[0047] In step S102, signal processing is performed based on the acquired drone RF signal and its SNR, and the RF signal is aligned. In some embodiments, the signal processing includes steps such as signal amplification, filtering and denoising, and the RF signal is aligned in time after processing.
[0048] In step S103, the radio frequency signal of each station after signal processing is input into a recurrent neural network LSTM, so as to use the LSTM to output the category to which the signal belongs, wherein the LSTM is combined with a time-series attention mechanism;
[0049] In step S104, the category to which the signal of each station outputted by LSTM belongs is weighted according to the corresponding SNR to determine the final signal recognition result, wherein the signal with a higher weight accounts for a larger proportion in the fusion result.
[0050] The embodiment of the present application increases the diversity of data and the robustness of signals by aggregating signal inputs from multiple sites, and the LSTM optimized through the temporal attention mechanism can deeply analyze and process the time series features in the signal, greatly improving the recognition accuracy.
[0051] In some embodiments, signal processing based on the acquired drone radio frequency signal and its SNR includes:
[0052] The acquired UAV radio frequency signal is amplified by an automatic gain control (AGC) amplifier. Signal amplification is the basis of preprocessing, and its purpose is to enhance the received weak signal to a sufficient amplitude. In the specific example of this application, the calculation of the gain G satisfies:
[0053]
[0054] Where V out and V in Represents the amplified and original signal voltages respectively. The key is to adjust the gain to the optimal level to avoid distortion or additional noise caused by over-amplification of the signal.
[0055] The amplified RF signal is low-pass filtered using a low-pass filter, and the transfer function of the low-pass filter satisfies:
[0056]
[0057] The low-pass filter has a cut-off frequency f c To attenuate the signal portion above a certain frequency.
[0058] The RF signal after low-pass filtering is denoised. Denoising clarifies the signal by reducing the noise in the signal. Applying a sliding average filter, denoising satisfies:
[0059]
[0060] Where M is the window size, which is used to reduce random noise by averaging a certain number of samples.
[0061] The denoised RF signals are aligned in time using a cross-correlation function. In some specific examples, the cross-correlation function is:
[0062]
[0063] The time delay τ between the signals can be estimated and adjusted to achieve the best synchronization effect.
[0064] In some embodiments, the radio frequency signal of each station after signal processing is input into a recurrent neural network LSTM, so that the category to which the LSTM output signal belongs includes:
[0065] The preprocessed RF signal IQ sequence is organized into the format (B, 1, H*W), where B represents the batch size, that is, the number of signal sequences processed at a time; 1 represents the number of channels of the signal sequence. For the IQ sequence, although it contains two components, I (in-phase) and Q (orthogonal), it is regarded as a composite signal channel here; H*W represents the length of the sequence, that is, the IQ sequence is regarded as a two-dimensional data with a height (H) of 1 and a width (W) of the sequence length.
[0066] By inputting the organized IQ sequence in a set format into LSTM, the signal can be recognized and classified efficiently. Figure 2 As shown, LSTM is a special recurrent neural network (RNN) that can process sequence data and capture long-term dependencies. Combined with the temporal attention mechanism, the model can focus more on the most important part of the sequence for the current task. The temporal attention mechanism enhances the model's ability to capture key information by assigning different weights to each time point in the sequence. The temporal attention mechanism of LSTM described in the embodiment of the present application is implemented through attention weights. For a sequence x=(x1, x2,…, x T ), the attention weight α t The attention score function is used to calculate and satisfy:
[0067]
[0068] Among them, s t-1 is the hidden state of LSTM at time point t-1, j t is the input at time point t, score is the function used to calculate the attention score, which is obtained by the dot product operation of the neural network. The attention score function depends on the current state of the sequence and the state at each time point.
[0069] After being processed by the LSTM, the output of each time point is weighted and summed according to its attention weight to form the final representation, and the final representation is classified through a fully connected layer, which is represented as:
[0070] y=softmax(W c (∑ t α t h t )+b c )
[0071] Among them, W c and b c are the weights and biases of the fully connected layer, and the softmax function is used to calculate the final classification probability. The model will output the category to which the signal belongs.
[0072] According to the classification results of each site, each site first independently makes a preliminary identification decision based on the received RF signal and measures the SNR of the signal. These preliminary decisions and SNR information are then transmitted to a central decision terminal. At the decision terminal, the present application adopts a weighted strategy to process information from different signal sources. Specifically, the SNR of each signal source is used as a basis for evaluating the quality and reliability of the signal, which determines its weight in the final decision. The higher the SNR of the signal source, the better the signal quality and the greater the corresponding weight, which means that the proportion and influence of the signal in the fusion result are also greater.
[0073] In some embodiments, the categories of signals of each station output by LSTM are weighted according to the corresponding SNRs to determine the final signal recognition result, including:
[0074] If there are signal recognition decisions from N different sites, each decision D i The SNR value corresponding to a signal source is SNR i , calculate the weight W of each decision according to the SNR value of each site i , to normalize the SNR value:
[0075]
[0076] The normalization formula ensures that the sum of all weights is 1, and the size of each weight is proportional to the SNR of its corresponding signal, thus reflecting the influence of signal quality in the final decision.
[0077] Then, according to the calculated weights of the signal sources of each site, the signal recognition result is determined by weighted average. The weighted fusion method of the embodiment of the present application allows the system to give priority to recognition decisions with high signal quality, thereby improving the overall recognition accuracy and reliability.
[0078] After calculating the weight of each signal source, the final signal identification result is determined by weighted average. In some embodiments, according to the calculated weights of the signal sources of each site, determining the signal identification result by weighted average includes:
[0079] The decision of each station is defined as the probability P of the signal belonging to a certain category. i (c), where c represents the possible category, then the weighted average category probability satisfies:
[0080]
[0081] According to the weighted average category probability, the final recognition category is determined based on the entropy weight adjustment mechanism.
[0082] In some embodiments, determining the final recognition category based on the entropy weight adjustment mechanism according to the weighted average category probability includes:
[0083] Using the entropy value H i To measure the uncertainty of each receiving station’s decision, satisfying:
[0084]
[0085] According to the entropy value H i Adjust the weight of each receiving station, the adjusted weight w i ′ for:
[0086]
[0087] This adjustment method weakens the impact of decisions with high uncertainty on the final fusion result, thereby improving the reliability of the fusion result. Further, the category with the highest weighted average probability is selected as the final recognition category.
[0088] c=arg max c P(c).
[0089] This design ensures that the most likely signal category is finally obtained by comprehensively considering the decision and signal quality of each site based on the fusion of multi-source signals. The above fusion method can effectively enhance the robustness and recognition accuracy of the system, especially when the signal quality of some sites is low or there is interference. By using the information of other high-quality signal sources, the uncertainty of the recognition results can be reduced and the performance of the overall system can be improved.
[0090] This application proposes a method for identifying UAV radio frequency (RF) signals based on multi-station fusion signal input and a recurrent neural network (LSTM) combined with a temporal attention mechanism. This application significantly improves the accuracy and efficiency of signal recognition. By aggregating signal inputs from multiple stations, not only the diversity of data and the robustness of signals are increased, but also the LSTM optimized by the temporal attention mechanism can deeply analyze and process the time series features in the signal, greatly improving the accuracy of recognition.
[0091] In addition, the preprocessing steps of this application, including signal amplification, filtering, denoising and synchronization processing, ensure the quality of the signal input to the LSTM, thereby improving processing efficiency and reducing the risk of misidentification. The decision-level fusion strategy further optimizes resource utilization by assigning corresponding weights to the signal-to-noise ratio (SNR) of different signal sources, ensuring the dominant role of high-quality signals in the recognition process.
[0092] Overall, the method of the present application achieves efficient and accurate identification of drone RF signals, has excellent adaptability and multi-task processing capabilities, and provides an efficient and reliable technical solution for drone monitoring and management.
[0093] The embodiment of the present application also proposes a UAV radio frequency signal recognition system based on multi-station fusion signals, which is characterized by comprising a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the UAV radio frequency signal recognition method based on multi-station fusion signals are implemented as described above.
[0094] It should be noted that in the various embodiments of the present application, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.
[0095] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.
[0096] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0097] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A method for identifying UAV radio frequency signals based on multi-station fusion signals, characterized in that: include: Deploy receiving stations at multiple geographical locations to capture drone RF signals and their signal-to-noise ratio (SNR) from different directions and distances; Perform signal processing based on the acquired drone RF signal and its SNR, and align the RF signal; Inputting the radio frequency signal of each station after signal processing into a recurrent neural network LSTM to utilize the LSTM to output the category to which the signal belongs, wherein the LSTM is combined with a temporal attention mechanism; The signal category of each station output by LSTM is weighted according to the corresponding SNR to determine the final signal recognition result, where the signal with a higher weight accounts for a greater proportion in the fusion result.
2. The method for identifying UAV radio frequency signals based on multi-station fusion signals according to claim 1, characterized in that: Signal processing based on the acquired drone RF signal and its SNR includes: The acquired UAV radio frequency signal is amplified using an automatic gain control (AGC) amplifier; The amplified radio frequency signal is low-pass filtered using a low-pass filter; Performing denoising on the low-pass filtered RF signal; The denoised RF signals are aligned in time using a cross-correlation function.
3. The method for identifying UAV radio frequency signals based on multi-station fusion signals according to claim 2, characterized in that: The RF signals of each station after signal processing are input into the recurrent neural network LSTM to use the LSTM output signal to belong to the following categories: Arrange the IQ sequence of the RF signal after signal processing into a set format (B, 1, H*W), where B represents the batch size, 1 represents the number of channels of the signal sequence, H*W represents the length of the sequence, and W is the width of the two-dimensional data; The sorted IQ sequence in a set format is input into the LSTM, wherein the temporal attention mechanism of the LSTM is implemented by the attention weight, and the attention weight α t The attention score function is used to calculate and satisfy: Among them, s t-1 is the hidden state of LSTM at time point t-1, h t is the input at time point t, score is the function used to calculate the attention score; After being processed by the LSTM, the output of each time point is weighted and summed according to its attention weight to form a final representation, and the final representation is classified through a fully connected layer.
4. The method for identifying UAV radio frequency signals based on multi-station fusion signals according to claim 3, characterized in that: The signal category of each station output by LSTM is weighted according to the corresponding SNR to determine the final signal recognition result including: Calculate the weight W of each decision based on the SNR value of each site i , to normalize the SNR value: The signal recognition result is determined by weighted averaging based on the calculated weights of the signal sources at each site.
5. The method for identifying UAV radio frequency signals based on multi-station fusion signals as claimed in claim 4, characterized in that: According to the calculated weights of the signal sources at each site, the signal identification results are determined by weighted average, including: The decision of each station is defined as the probability P of the signal belonging to a certain category. i (c), where c represents the possible category, then the weighted average category probability satisfies: According to the weighted average category probability, the final recognition category is determined based on the entropy weight adjustment mechanism.
6. The method for identifying UAV radio frequency signals based on multi-station fusion signals according to claim 5, characterized in that: According to the weighted average category probability, the entropy weight-based adjustment mechanism determines the final recognition category, including: Using the entropy value H i To measure the uncertainty of each receiving station’s decision, satisfying: According to the entropy value H i Adjust the weight of each receiving station, the adjusted weight w i ′ for: By selecting the category with the highest weighted average probability as the final recognition category c=arg max c P(c).
7. The method for identifying UAV radio frequency signals based on multi-station fusion signals according to claim 1, characterized in that: Deploying receiving stations at multiple geographical locations includes arranging the receiving stations to cover the possible activity ranges and flight paths of the drones.
8. A UAV radio frequency signal recognition system based on multi-station fusion signals, characterized in that: It includes a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method for identifying unmanned aerial vehicle radio frequency signals based on multi-station fusion signals as described in any one of claims 1 to 7 are implemented.