A post-processing method, device and equipment for drone detection and identification based on machine learning

Abstract

The present application discloses a post-processing method, device and equipment for drone detection and identification based on machine learning, the method comprising: automatically collecting drone signals, performing signal transformation on the collected continuous signal segments, the signal transformation including changing the transmission frequency, etc., inputting the transformed signal into a neural network for judgment to obtain a judgment result of a single signal segment, and performing a joint judgment of the judgment result of the continuous signal using a sliding window to obtain a final judgment result. The present application realizes joint judgment of multiple scenarios through signal transformation and sliding windows of multiple signal segments. It reduces the misjudgment caused by a single signal segment in a single scene, so that the effect of each joint judgment in multiple scenarios is roughly the same.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence and signal processing, and specifically to a post-processing method, device and equipment for drone detection and identification based on machine learning. Background Art

[0002] In recent years, the drone industry has developed rapidly, and the trend of commercial, civilian, and even military use has become increasingly obvious, including: legal aerial photography, cargo transportation, drone performances, drone competitions, smart agriculture, plant protection, disaster relief, wildlife observation, surveying and mapping, news reporting, power inspections, etc. At the same time, there are also some illegal applications, such as illegal intrusion, sneak photography, throwing illegal objects, etc., and there are more and more public safety hazards. Although many countries are paying more and more attention to the safety of drones and have gradually formulated corresponding laws and regulations to regulate them, the above-mentioned illegal applications, as well as some irregular operations and drone out of control, cannot be effectively controlled by legal means alone. Therefore, professional equipment is needed to manage them, including detection and defense. Among them, drone detection is a basic requirement for drone management.

[0003] Currently, commonly used drone detection methods include active radar detector detection, infrared camera detection, visual detection, sound-based auditory detection, and detection based on drone-transmitted radio signals. Patent CN202110487167.5 discloses a drone detection method for different signal systems. Specifically: a signal receiving device for a specific system signal is used to output its full-band spectrum data and IQ data stream; a power spectrum cancellation method is used for noise reduction processing, and then random noise signals are removed; feature parameters are extracted, and they are matched, classified, and identified with parameters in the database; WiFi system drone signals are detected and identified, and the MAC address of the WiFi system drone signal obtained by parsing is matched and identified with the known drone manufacturer MAC address.

[0004] Machine learning mainly extracts the features of input signals by training neural networks. Figure 1 This is a simple diagram of machine learning, where circles represent values ​​and arrows represent operations. The essence of the neural network training process is to find the optimal operation parameters (i.e. Figure 1 The present invention is based on machine learning technology, generates an optimal neural network by training the neural network, and then uses the network to detect and identify the drone.

[0005] In terms of detection of radio signals transmitted by drones, drone identification methods based on data enhancement have emerged. Patent CN201910930192.9 relates to a method for identifying drone remote control signals based on frequency domain features, which performs mixing, phase detection, sampling, and channelization processing on received drone signals to obtain frequency measurement outputs of multiple channels, thereby realizing detection and identification of drone remote control signals.

[0006] However, after the training of the neural network based on the radio signal of the drone is completed, that is, when the drone identification device storing the trained drone identification model is used to perform the actual drone identification process, the identification effect is still not ideal due to the following two reasons: First, when the drone signal is actually identified, the collection scene of the collected drone signal is often limited. It is difficult to accurately identify the drone signal when identifying the drone signal collected in the limited scene. Second, when using the decision network to judge the drone signal fragment, relying on a single signal fragment as the final decision result, this often does not get the correct result and has poor reliability. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a post-processing method, device and equipment for drone detection and identification based on machine learning. In drone identification, there is no need to actually provide drone signals in many scenarios. It is only necessary to collect drone signals in conventional scenarios, and then obtain drone signals in multiple scenarios through signal transformation. Each signal segment is sent to a neural network for identification and judgment, and the judgment result of a single signal segment is obtained. The judgment result of the continuous signal is jointly judged by a sliding window to obtain the final judgment result.

[0008] Machine learning mainly extracts the features of input signals by training neural networks. The present invention obtains a drone recognition model by training the neural network, and then detects and identifies drones through the network.

[0009] The specific technical solutions of the present invention are as follows:

[0010] In a first aspect, the present application provides a post-processing method for drone detection and identification based on machine learning, the method comprising:

[0011] Step 1: Use the drone signal acquisition device to automatically collect the drone's communication signals.

[0012] Step 2: Segment the collected drone signal to obtain multiple continuous signal segments with fixed length.

[0013] Step 3, performing signal transformation on the signal segment obtained in step 2, inputting the transformed signal into a decision network for decision, and obtaining a decision result of a single signal segment.

[0014] Step 4: perform joint decision of the sliding window according to the decision results of the continuous segments.

[0015] In the step 1, the communication signals of the drone collected by the drone signal collection device include communication signals used when the drone communicates with the ground and the control end, and also include environmental signals.

[0016] In the step 1, after the signal is collected, the collected signal is added to the signal library.

[0017] In the step 3, signal transformation is performed on the signal segment, and the signal transformation includes one or more of the following operations: changing the transmission frequency of the signal, changing the signal-to-interference ratio of the signal, and changing the signal-to-noise ratio of the signal.

[0018] A wireless channel refers to a data signal transmission channel that uses wireless signals as the transmission medium. A drone usually has multiple wireless channels, corresponding to multiple transmission frequencies. By changing the transmission frequency of the signal, the signal in each wireless channel of the drone can be simulated.

[0019] The method for changing the transmission frequency of a signal includes: assuming that the original signal is {x[0], x[1], x[2], ..., x[N-1]}, where N is the length of the wireless signal. Assuming that the channel frequency of the original signal is f0, the set of all channel frequencies of the signal is {f0, f1, ..., f M}, where M is the total number of channels. Assume that the signal frequency point to be transformed in the data enhancement parameter is f m , the sampling frequency is f s , then the signal is transformed into:

[0020]

[0021] The signal-to-interference ratio conversion can simulate the actual situation that the receiving end in the wireless communication system receives both the useful signal and the interference signal when receiving the signal. The signal-to-interference ratio conversion is specifically:

[0022] Assume that the required signal-to-interference ratio is α, and the interference signal is {I[0],I[1],I[2],...,I[N-1]}, then

[0023]

[0024] in:

[0025] Then, the signal is transformed into:

[0026]

[0027] The signal-to-noise ratio conversion can simulate the attenuation effect of different degrees of attenuation due to the transmission distance when the wireless signal reaches the receiving end. The signal-to-noise ratio conversion is specifically:

[0028] Assume the noise power is P noise , the required signal-to-noise ratio is β, then:

[0029]

[0030] in:

[0031] Then, the signal is transformed into:

[0032]

[0033] in, The power is P noise is a circularly symmetric zero-mean complex Gaussian random variable.

[0034] After the signal transformation in step 3, the signal to be determined is transformed into signals under multiple scenarios rather than the signal of a single scenario, so that the joint determination effect each time is roughly the same.

[0035] In step 4, the joint decision is specifically as follows: set the sliding window size to K, count the number of signal segments determined to contain the target drone in K consecutive signal segments, and record it as K pos If K pos If the value is greater than the threshold, the final judgment result is that there is a drone in the surrounding environment; otherwise, the final judgment result is that there is no drone in the surrounding environment.

[0036] The joint judgment of multiple signal segments in step 4 reduces the misjudgment caused by a single signal segment.

[0037] In a second aspect, the present application provides a post-processing device for drone detection and identification based on machine learning, the device comprising:

[0038] The UAV signal acquisition module automatically collects the communication signals of the UAV.

[0039] The signal segment module is connected to the UAV signal acquisition module to segment the collected UAV communication signal to obtain multiple continuous signal segments.

[0040] The signal conversion module is connected to the signal segment module and performs signal conversion on the plurality of continuous signal segments, and uses the converted data as the input of the decision network.

[0041] The decision network module is connected to the signal conversion module and is used to make a decision on the transformed signal to obtain a decision result of a single signal segment.

[0042] The sliding window joint decision module is connected to the decision network module and performs a sliding window joint decision according to the decision results of the continuous segments.

[0043] The communication signals of the drone collected by the drone signal collection module include the communication signals used by the drone to communicate with the ground and the control terminal, and also include environmental signals. The method and steps of the automatic collection are the same as those described above and will not be repeated here.

[0044] The signal conversion module performs data conversion on the collected signal, and includes one or more of the following modules: a transmission frequency conversion module, a signal-to-interference ratio conversion module, and a signal-to-noise ratio conversion module.

[0045] The transmission frequency conversion module, the signal-to-interference ratio conversion module, and the signal-to-noise ratio conversion module respectively realize the change of the transmission frequency of the signal, the change of the signal-to-interference ratio of the signal, and the change of the signal-to-noise ratio of the signal as mentioned above, which will not be repeated here.

[0046] The specific joint decision is as follows: set the sliding window size to K, count the number of signal segments that are judged to contain the target drone in K consecutive signal segments, and record it as K pos If K pos If the value is greater than the threshold, the final judgment result is that there is a drone in the surrounding environment; otherwise, the final judgment result is that there is no drone in the surrounding environment.

[0047] The joint judgment of multiple signal segments reduces the misjudgment caused by a single signal segment.

[0048] In a third aspect, the present application also provides a post-processing device for drone detection and identification based on machine learning, wherein the device includes a memory and a processor.

[0049] The memory is used to store program codes and transmit the program codes to the processor.

[0050] The processor is used to execute the post-processing method for drone detection and identification and the drone detection method described in the first aspect according to the instructions in the program code.

[0051] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the post-processing method for drone detection and identification based on machine learning and the steps of the drone detection method described in the first aspect.

[0052] The beneficial effects of this application include:

[0053] 1. The present application provides a multi-signal segment judgment mechanism after signal transformation, so that the neural network has roughly the same drone recognition effect in each scenario, and can achieve a high judgment accuracy, avoiding misjudgment caused by the judgment results of a single scenario and a single signal.

[0054] 2. In the data conversion link, the present application performs arbitrary signal conversion on the signal to be processed, and converts the signal to be judged into signals in multiple scenarios rather than signals in a single scenario, so that the joint judgment effect each time is roughly the same. For example, by changing the transmission frequency of the signal, the signal in each wireless channel of the drone can be simulated. Using signal-to-interference ratio conversion for data conversion can simulate the actual situation that the receiving end in a wireless communication system receives both useful signals and interference signals when receiving signals. Using signal-to-noise ratio conversion for data conversion can simulate the attenuation effect of different degrees of attenuation due to the transmission distance when the wireless signal transmission reaches the receiving end. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0056] Figure 1 It is a structural diagram of a neural network model in a drone detection device based on machine learning;

[0057] Figure 2 A flowchart of a post-processing method for drone detection and identification based on machine learning provided in an embodiment of the present application;

[0058] Figure 3 A schematic diagram of the structure of a post-processing device for drone detection and identification based on machine learning provided in an embodiment of the present application;

[0059] Figure 4 A schematic diagram of a sliding window of a post-processing device for drone detection and identification based on machine learning provided in an embodiment of the present application.

[0060] Figure 5 A schematic diagram of the structure of a drone detection device based on machine learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0062] In the current drone detection based on drone radio signals, after the training of the neural network based on drone radio signals is completed, that is, when the drone identification device storing the trained drone identification model is used to perform actual drone identification, the identification effect is still not ideal due to the following two reasons: First, when actually performing drone signal identification, the collection scenes of the collected drone signals are often limited, and it is difficult to accurately identify the drone signals collected in limited scenes. Second, when using the decision network to judge the drone signal fragment, relying on a single signal fragment for judgment as the final judgment result, this often does not obtain the correct result and has poor reliability.

[0063] To this end, the present application provides a post-processing method, device and equipment for drone detection and identification based on machine learning. In drone identification, there is no need to actually provide drone signals in many scenarios. It is only necessary to collect drone signals in conventional scenarios, and then obtain drone signals in multiple scenarios through signal transformation. Each signal segment is sent to the neural network for identification and judgment, and the judgment result of a single signal segment is obtained. The judgment result of the continuous signal is jointly judged by a sliding window to obtain the final judgment result.

[0064] The following specifically introduces an embodiment of a data enhancement method in signal machine learning and a drone detection method provided by this application, with reference to Figure 2 , is a flow chart of a method provided in an embodiment of the present application, the method specifically includes:

[0065] S101: Automatically collect the communication signal of the drone using the drone signal collection device.

[0066] In the embodiment of the present application, the communication signal of the drone collected by the drone signal collection device includes the communication signal used when the drone communicates with the ground and the control end, and also includes the environmental signal. In addition, after the signal is collected, the collected signal is added to the signal library.

[0067] S102: Using a signal segment module, the collected drone signal is segmented to obtain a plurality of continuous signal segments with fixed lengths.

[0068] S103: Using a signal transformation module, perform signal transformation on the plurality of continuous signal segments, and use the transformed data as input of a decision network.

[0069] S104: using a decision network module to make a decision on the transformed signal to obtain a decision result of a single signal segment.

[0070] S105: Using the sliding window joint decision module, perform sliding window joint decision according to the decision results of the continuous segments to obtain a final decision result.

[0071] In S103, signal transformation is performed on the signal segment, including one or more of the following operations: changing the transmission frequency of the signal, changing the signal-to-interference ratio of the signal, changing the signal-to-noise ratio of the signal, etc.

[0072] A wireless channel refers to a data signal transmission channel that uses wireless signals as the transmission medium. A drone usually has multiple wireless channels. Channelization conversion is to convert the wireless signal on a certain channel into the signal on any other channel.

[0073] The method for changing the transmission frequency of the signal is as follows: Assume that the original signal is {x[0], x[1], x[2], ..., x[N-1]}, where N is the length of the wireless signal. Assume that the channel frequency of the original signal is f0, and the set of all channel frequencies of the signal is {f0, f1, ..., f M}, where M is the total number of channels. Assume that the signal frequency point to be transformed in the data enhancement parameter is f m , the sampling frequency is f s , then the signal is transformed into:

[0074]

[0075] The signal-to-interference ratio conversion is to simulate the actual situation that the receiving end in the wireless communication system receives both the useful signal and the interference signal when receiving the signal. The signal-to-interference ratio conversion is specifically:

[0076] Assume that the required signal-to-interference ratio is α, and the interference signal is {I[0],I[1],I[2],...,I[N-1]}, then

[0077]

[0078] in:

[0079] Then, the signal is transformed into:

[0080]

[0081] The signal-to-noise ratio conversion is to simulate the attenuation effect of different degrees of attenuation due to the transmission distance when the wireless signal reaches the receiving end. The signal-to-noise ratio conversion is specifically:

[0082] Assume the noise power is P noise , the required signal-to-noise ratio is β, then:

[0083]

[0084] in:

[0085] Then, the signal is transformed into:

[0086]

[0087] in, The power is P noise is a circularly symmetric zero-mean complex Gaussian random variable.

[0088] The specific joint decision is as follows: set the sliding window size to K, count the number of signal segments that are judged to contain the target drone in K consecutive signal segments, and record it as K pos If K pos If the value is greater than the threshold, the final judgment result is that there is a drone in the surrounding environment; otherwise, the final judgment result is that there is no drone in the surrounding environment.

[0089] The present invention transforms the signal to be judged into signals in multiple scenarios rather than signals in a single scenario, so that the joint judgment effect each time is roughly the same. At the same time, the joint judgment of multiple signal fragments is adopted to reduce the misjudgment caused by a single signal fragment. The present invention organically combines the two and achieves better beneficial effects: First, if only signal transformation is performed without sliding window joint judgment, that is, a single signal fragment is judged on the multi-scenario signal after the signal transformation, it is only equivalent to providing a multi-scenario signal to be judged, but the specific judgment is still the judgment result of the single fragment of the selected single scene; second, if no signal transformation is performed, only a sliding window joint judgment is performed, that is, a sliding window joint judgment is performed on the signal of a single scene, then only the randomness problem caused by a single judgment of a single scene signal is avoided, and a comprehensive judgment under multiple scenarios cannot be achieved.

[0090] Corresponding to the above method embodiments, the present application also provides a post-processing device for drone detection and identification based on machine learning, see Figure 3 , the device comprises:

[0091] The signal segment module is connected to the UAV signal acquisition module to segment the collected UAV communication signal to obtain multiple continuous signal segments.

[0092] The signal conversion module is connected to the signal segment module and performs signal conversion on the plurality of continuous signal segments, and uses the converted data as the input of the decision network.

[0093] The decision network module is connected to the signal conversion module and is used to make a decision on the transformed signal to obtain a decision result of a single signal segment.

[0094] The sliding window joint decision module is connected to the decision network module, and performs a sliding window joint decision according to the decision results of the continuous segments to obtain a final decision result.

[0095] The data conversion module performs data conversion on the signal segments, and includes one or more of the following modules: a transmission frequency conversion module, a signal-to-interference ratio conversion module, a signal-to-noise ratio conversion module, etc.

[0096] The transmission frequency conversion module, signal-to-interference ratio conversion module, and signal-to-noise ratio conversion module respectively implement the transmission frequency conversion, signal-to-interference ratio conversion, and signal-to-noise ratio conversion as described above based on the data conversion parameters, and will not be repeated here.

[0097] The embodiment of the present invention also provides a sliding window schematic diagram of a post-processing device for drone detection and identification based on machine learning, see Figure 4 As shown:

[0098] Set the sliding window size to K, and count the number of signal segments that are determined to contain the target drone among the K consecutive signal segments, denoted as K pos If K pos If the value is greater than the threshold, the final judgment result is that there is a drone in the surrounding environment; otherwise, the final judgment result is that there is no drone in the surrounding environment.

[0099] The present invention also provides a machine learning-based drone detection device, see Figure 5 As shown, this may include:

[0100] Processor 501, memory 502, input device 503 and output device 504. The number of processors 501 in the drone detection device based on machine learning can be one or more. Figure 4 In some embodiments of the present invention, the processor 501, the memory 502, the input device 503 and the output device 504 may be connected via a bus or other means, wherein: Figure 4 The example of connecting through bus is taken in the following.

[0101] The memory 502 can be used to store software programs and modules. The processor 501 executes various functional applications and data processing of the drone detection device based on machine learning by running the software programs and modules stored in the memory 502. The memory 502 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc. In addition, the memory 502 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. The input device 503 may be used to receive input digital or character information, and to generate signal input related to user settings and function control of the drone detection device based on machine learning.

[0102] Specifically in this embodiment, the processor 501 will load the executable files corresponding to the processes of one or more applications into the memory 502 according to the following instructions, and the processor 501 will run the applications stored in the memory 502, thereby realizing the various functions of the above-mentioned drone detection method based on machine learning.

[0103] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0104] It should be noted that, in this article, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus.

[0105] The above is a detailed introduction to the drone detection and identification method, device and equipment based on machine learning provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present application, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A post-processing method for drone detection and identification based on machine learning, characterized in that: The method comprises: Automatically collect signals from drones; Segment the collected drone signals to obtain multiple continuous signal segments with fixed lengths; Performing signal transformation on the plurality of continuous signal segments, and using the transformed data as input of a decision network; The signal transformation includes: changing the transmission frequency of the signal; simulating the signals in each wireless channel of the drone by changing the transmission frequency of the signal; Making a decision on the transformed signal to obtain a decision result of a single signal segment; and A joint decision of the sliding window is performed based on the decision results of the continuous segments to obtain the final decision result.

2. The method according to claim 1, characterized in that: The signal transformation includes one or more of the following operations: changing the transmission frequency of the signal, changing the signal-to-interference ratio of the signal, and changing the signal-to-noise ratio of the signal.

3. The method according to claim 1 or 2, characterized in that: The method for changing the transmission frequency of a signal includes: assuming that the original signal is {x[0], x[1], x[2], ..., x[N-1]}, where N is the length of the wireless signal, assuming that the channel frequency of the original signal is f0, and the set of all channel frequencies of the signal is {f0, f1, ..., f M }, where M is the total number of channels, assuming that the signal frequency point to be transformed in the data enhancement parameter is f m , the sampling frequency is f s , then the signal is transformed into:

4. The method according to claim 2, characterized in that: The method for changing the signal-to-interference ratio of a signal includes: the signal input to the signal-to-interference ratio conversion is y1[n], the target signal-to-interference ratio of the signal-to-interference ratio conversion is α, the interference signal is {I[0], I[1], I[2], ..., I[N-1]}, and the signal-to-interference ratio conversion converts the signal y1[n] into y2[n]: in, x[n] is the original signal, N is the length of the wireless signal, n=0, 1, 2, ..., N-1.

5. The method according to claim 2, characterized in that: The method for changing the signal-to-noise ratio of a signal includes: the signal input to the signal-to-noise ratio conversion is y2[n], the power of the noise is P noise , the target signal-to-noise ratio is β, and the signal-to-noise ratio transformation transforms the signal y2[n] into y3[n]: in, The power is P noise is a cyclically symmetric zero-mean complex Gaussian random variable, N is the length of the wireless signal, n=0, 1, 2, ..., N-1.

6. The method according to claim 1 or 2, characterized in that: The sliding window joint judgment is as follows: set the sliding window size to K, count the number of signal segments Kpos judged to contain the target UAV in K consecutive signal segments, if Kpos is greater than the threshold, the judgment result is that the UAV exists, otherwise, the judgment result is that the UAV does not exist.

7. A post-processing device for drone detection and identification based on machine learning, characterized in that: The device comprises: The signal segment module segments the collected UAV communication signal to obtain multiple continuous signal segments; A signal transformation module, connected to the signal segment module, performs signal transformation on the plurality of continuous signal segments, and uses the transformed data as input of the decision network; The signal conversion module includes: a transmission frequency conversion module; by changing the transmission frequency of the signal, the signal in each wireless channel of the drone is simulated; A decision network module, connected to the signal conversion module, is used to make a decision on the converted signal to obtain a decision result of a single signal segment; and The sliding window joint decision module is connected to the decision network module, and performs a sliding window joint decision according to the decision results of the continuous segments to obtain a final decision result.

8. The device according to claim 7, characterized in that The specific joint decision of sliding window is: Set the sliding window size to K, and count the number of signal segments Kpos judged to contain the target drone in K consecutive signal segments. If Kpos is greater than the threshold, the result is that the drone exists, otherwise, the result is that the drone does not exist.

9. A drone detection device based on machine learning, characterized in that: The device comprises a memory and a processor; The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the method according to any one of claims 1 to 6 according to the instructions in the program code.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images