Unmanned aerial vehicle discovery method based on RF
By combining active radio frequency scanning and spectrum analysis with deep learning technology, the problems of accuracy and efficiency in drone detection have been solved, enabling rapid drone detection and identification in complex environments.
Patent Information
- Application Number
- CN202511419516.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-30
- Publication Date
- 2026-01-09
AI Technical Summary
Existing technologies lack drone detection solutions that combine active radio frequency transmission, spectrum analysis, and deep learning, resulting in insufficient accuracy and efficiency in drone detection in complex environments, especially in the poor identification of hibernating or encrypted drones.
By employing active radio frequency scanning to transmit signals, combined with spectrum analysis and deep learning technologies, and through radio frequency reception, signal preprocessing, time-spectrum analysis, and convolutional neural network identification of UAV characteristics, efficient and accurate detection and identification of UAVs can be achieved.
It can quickly detect drones in various scenarios, improves robustness and recognition accuracy in complex electromagnetic environments, and supports rapid identification of new drone models.
Smart Images
Figure CN121308901A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) detection and identification technology, and specifically relates to a UAV detection method based on radio frequency (RF). This method achieves effective detection and identification of UAV targets in different environments through active transmission of radio frequency signals, spectrum analysis, and deep learning feature recognition. Background Technology
[0002] With the rapid development of drone technology, drones have been widely used in military reconnaissance, aerial photography, logistics, and other fields, but they also bring many security risks. For example, in environments such as civilian airports, important locations, and border security, unauthorized drone intrusions may pose risks to air security and information leaks. Therefore, accurately and promptly detecting and identifying drones in the surrounding space has become an urgent technical problem to be solved.
[0003] Currently, drone detection methods mainly include optical / infrared imaging, acoustic detection, radar detection, and radio frequency signal detection. Among them: Optical / infrared imaging relies on cameras to scan targets visually or with infrared images, and is easily limited by conditions such as lighting and weather. Acoustic detection uses a microphone array to pick up the drone's acoustic signature, but its detection range is limited and it is sensitive to environmental noise. Radar detection uses the principle of radio wave reflection to detect targets and can detect non-cooperative targets. However, traditional radar systems are large, expensive, and have strict requirements on the radar cross-section of small UAVs. Radio frequency (RF) detection primarily identifies drones by analyzing the radio signal characteristics of their control or image transmission links. Since drones typically operate in specific frequency bands (such as 2.4 GHz, 5.8 GHz, etc.), this method can effectively identify some drones in passive situations. However, passive RF detection relies on signals actively emitted by the drone, making it less reliable for drones in hibernation or those employing spectrum hopping or encryption. Furthermore, passive detection is sensitive to background radio interference, making it prone to missed detections or false alarms.
[0004] Deep learning technology can extract features from large amounts of data to perform tasks such as classification, regression, and generation. Applying deep learning technology to drone detection is a research direction that has emerged in recent years.
[0005] Currently, there is no complete drone detection solution that combines active radio frequency transmission, spectrum analysis, deep learning, and feature matching. Summary of the Invention
[0006] To address the problems existing in the prior art, the purpose of this invention is to provide a method for detecting drones based on active radio frequency detection. By actively transmitting wireless signals and combining technologies such as spectrum analysis and deep learning, this method can achieve efficient and accurate detection and identification of drones within a target area.
[0007] The present invention adopts the following technical solution: A method for detecting drones based on radio frequency (RF) includes the following steps: Step S1: Radio frequency scanning and transmission; Set up an RF transmitter to actively transmit RF scanning signals within a predetermined frequency range.
[0008] This step involves actively transmitting signals to detect non-cooperative drone targets, providing a signal source for subsequent passive reception and analysis.
[0009] Step S2: Signal reception and preprocessing; A radio frequency receiver is used to receive signals from the environment within the target area, and the received signals are preprocessed to obtain baseband digital signals.
[0010] Step S3: Spectrum analysis and feature extraction; Perform time-spectrum analysis on the preprocessed digital signal; extract high-dimensional features from the spectrogram / time-spectrum; use the extracted features as input feature vectors and provide them to the neural network in step S4; alternatively, the spectrogram / time-spectrum image can be directly provided to the neural network in step S4.
[0011] Step S4: Deep Learning Recognition The feature vectors or time-spectrum maps obtained in step S3 are input into the convolutional neural network (CNN) model for UAV feature recognition.
[0012] Step S5: Feature matching and decision-making.
[0013] The low-dimensional feature vectors output by the CNN model are matched with a pre-built drone feature database to complete the final detection and judgment.
[0014] Beneficial effects Through the above technical solutions, this invention can quickly detect drones in various scenarios (such as open areas, urban environments, or complex interference backgrounds) and supports rapid identification of new drone models. Active RF signal transmission can proactively detect non-cooperative targets, while deep learning enhances robustness and recognition accuracy in complex electromagnetic environments. Attached Figure Description
[0015] Figure 1 This is a flowchart of the UAV detection method of the present invention; Detailed Implementation The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments, but the present invention is not limited to the following embodiments.
[0016] A method for detecting drones based on active radio frequency detection includes the following steps: (e.g.) Figure 1 ) Step S1, Radio Frequency Scanning Transmission Set up an RF transmitter to actively transmit RF scanning signals within a predetermined frequency range.
[0017] Specifically, the radio frequency scanning signal transmission frequency is selected from two frequency bands commonly used by UAVs: 2.4GHz / 2.48GHz or 5.72GHz / 5.85GHz. The target area is scanned across the entire frequency band by linear frequency modulation (LFM) sweep signal or pulse signal.
[0018] Furthermore, the bandwidth and transmission power of the radio frequency scanning signal can be set as needed to cover possible UAV communication frequency bands and ensure a certain detection distance.
[0019] Furthermore, the radio frequency transmitter employs a wide-beam transmitting antenna with a main lobe width of 60°.
[0020] Specifically, the radio frequency scanning signal continuously sweeps the frequency band of 2.4 GHz with a bandwidth of 100 MHz, and each sweep period is less than 1 second.
[0021] This step involves actively transmitting signals to detect non-cooperative drone targets, providing a signal source for subsequent passive reception and analysis.
[0022] Step S2: Signal reception and preprocessing Use an RF receiver to receive signals from the environment within the target area, including: echoes of transmitted signals reflected back from the drone, and control or image transmission signals transmitted by the drone itself.
[0023] The received signal is preprocessed to obtain the baseband digital signal.
[0024] Specifically, the received analog radio frequency signal first enters an analog-to-digital converter (ADC) for sampling and conversion into a digital signal. It then undergoes digital down-conversion and bandpass filtering to remove irrelevant noise and interference signals. The preprocessed baseband digital signal contains characteristic information that may be generated by the drone.
[0025] Step S3: Spectrum Analysis and Feature Extraction Perform time-frequency analysis on the preprocessed digital signal. Use Fast Fourier Transform (FFT) or Short-Time Fourier Transform (STFT) to convert the time-domain signal into a frequency-domain or time-frequency-domain representation.
[0026] High-dimensional features are extracted from the spectrogram / time-spectrum. Specifically, the possible modulation characteristics of the UAV (e.g., FDM, OFDM, etc.) and spectral peaks can be observed. Based on this, signal parameters reflecting the characteristics of the UAV are extracted, such as the main carrier frequency, bandwidth characteristics, modulation line position, echo delay, etc.
[0027] The extracted features are used as input feature vectors and provided to the neural network in step S4; alternatively, the spectrogram / time-spectrum image can be directly provided to the neural network in step S4.
[0028] Step S4: Deep Learning Recognition The feature vectors or time-spectrum maps obtained in step S3 are input into the convolutional neural network (CNN) model for UAV feature recognition.
[0029] The Convolutional Neural Network (CNN) model includes an input layer, multiple convolutional layers and pooling layers, and one or more fully connected layers. The convolutional layers use several small convolutional kernels (e.g., 3×3) to extract features from the temporal spectrogram. The pooling layers are used for downsampling to extract translation-invariant features. The fully connected layers map high-level features to the final classification output. The structural parameters of the CNN model (such as the number of convolutional kernels, the number of layers, and the pooling method) are optimized according to the specific application scenario.
[0030] During the training phase, the system uses a large labeled dataset containing signals from different drone models to supervise the learning of the CNN model, enabling the model to learn the radio frequency characteristics of various drones and their differences. The training process adjusts the network weights by optimizing the loss function (such as cross-entropy).
[0031] During the inference phase, the CNN model output includes: the probability of the presence of a drone, the drone type classification result, or a low-dimensional feature vector.
[0032] By introducing a CNN network, the system exhibits good robustness and high accuracy against complex wireless interference.
[0033] Step S5: Feature Matching and Decision The low-dimensional feature vectors output by the CNN model are matched with a pre-built drone feature database to complete the final detection and judgment.
[0034] The feature database contains radio frequency features of known drone models, which are high-level feature vectors extracted by CNN or manually defined features.
[0035] The matching process can use Euclidean distance, cosine similarity, or other metrics to calculate the similarity between the input features and known features in the database. When the similarity exceeds a set threshold, a drone is detected, and specific information is provided based on the matched model; otherwise, no drone is detected or the drone detection failed. Finally, the system outputs the decision result of whether a drone was detected and can trigger an alarm or log for subsequent processing.
[0036] The above description is merely a description of preferred embodiments of this application and is not intended to limit the scope of this application in any way. Any changes or modifications made by those skilled in the art based on the above-disclosed technical content should be considered as equivalent and valid embodiments and fall within the scope of protection of the technical solution of this application.
Claims
1. A method for detecting unmanned aerial vehicles (UAVs) based on radio frequency (RF), characterized in that, Includes the following steps: Step S1: Radio frequency scanning and transmission; Set up an RF transmitter to actively transmit RF scanning signals within a predetermined frequency range; Step S2: Signal reception and preprocessing; The radio frequency receiver is used to receive signals from the environment in the target area, and the received signals are preprocessed to obtain baseband digital signals; Step S3: Spectrum analysis and feature extraction; Perform time-spectrum analysis on the preprocessed digital signal; extract high-dimensional features from the spectrogram / time-spectrum; use the extracted features as input feature vectors and provide them to the neural network in step S4; alternatively, the spectrogram / time-spectrum image can be directly provided to the neural network in step S4. Step S4: Deep learning recognition; Input the feature vector or time-spectrum graph obtained in step S3 into the convolutional neural network (CNN) model for UAV feature recognition; Step S5: Feature matching and decision-making; The low-dimensional feature vectors output by the CNN model are matched with a pre-built drone feature database to complete the final detection and judgment.
2. The method for detecting unmanned aerial vehicles based on radio frequency according to claim 1, characterized in that, In step S1, the radio frequency scanning signal transmission frequency is selected from two frequency bands: 2.4GHz / 2.48GHz or 5.72GHz / 5.85GHz. The target area is scanned across the entire frequency band by a linear frequency modulation sweep signal or a pulse signal.
3. The method for detecting unmanned aerial vehicles based on radio frequency according to claim 2, characterized in that, In step S1, the radio frequency scanning signal continuously sweeps the frequency band of 2.4 GHz with a bandwidth of 100 MHz, and each sweep period is less than 1 second.
4. The method for detecting unmanned aerial vehicles based on radio frequency according to claim 1, characterized in that, Step S2 is as follows: Use an RF receiver to receive signals from the environment within the target area, including: echoes of transmitted signals reflected back from the drone, and control or image transmission signals transmitted by the drone itself. The received signal is preprocessed to obtain a baseband digital signal. Specifically, the received analog radio frequency signal first enters the analog-to-digital converter for sampling and conversion into a digital signal, and then undergoes digital down-conversion and bandpass filtering to filter out irrelevant noise and interference signals. The baseband digital signal obtained after preprocessing contains characteristic information that may be generated by the UAV.
5. The method for detecting unmanned aerial vehicles based on radio frequency according to claim 1, characterized in that, Step S3 is as follows: Perform time-frequency analysis on the preprocessed digital signal; use Fast Fourier Transform or Short-Time Fourier Transform to convert the time-domain signal into frequency domain or time-frequency domain representation; Extract high-dimensional features from spectrograms / time-spectrums; extract signal parameters reflecting UAV characteristics; The extracted features are used as input feature vectors and provided to the neural network in step S4; alternatively, the spectrogram / time-spectrum image can be directly provided to the neural network in step S4.
6. The method for detecting unmanned aerial vehicles based on radio frequency according to claim 1, characterized in that, Step S4 is as follows: Input the feature vector or time-spectrum graph obtained in step S3 into the convolutional neural network (CNN) model for UAV feature recognition; The convolutional neural network model includes an input layer, multiple convolutional layers and pooling layers, and one or more fully connected layers. The convolutional layers use several small-sized convolutional kernels to extract features from the temporal spectrogram, the pooling layers are used for downsampling to extract features with translation invariance, and the fully connected layers map high-level features to the final classification output. During the training phase, the system uses a large number of labeled datasets containing signals from different drone models to supervise the learning of the CNN model, enabling the model to learn the radio frequency characteristics of various drones and their differences. The training process adjusts the network weights by optimizing the loss function; During the inference phase, the CNN model output includes: the probability of the presence of a drone, the drone type classification result, or a low-dimensional feature vector.
7. The method for detecting unmanned aerial vehicles based on radio frequency according to claim 1, characterized in that, Step S5 specifically involves matching the low-dimensional feature vector output by the CNN model with the pre-built UAV feature database to complete the final detection and judgment. The feature database contains radio frequency features of known drone models, which are high-level feature vectors extracted by CNN or manually defined features; The matching process calculates the similarity between the input features and each known feature in the database. When the similarity exceeds a set threshold, it is determined that a drone has been detected, and specific information is given based on the matched model. Otherwise, it is determined that there is no drone or the drone detection failed; finally, the system outputs the decision result of whether the drone was detected, and may trigger an alarm or log for subsequent processing.