Decision tree and radial basis SVM linkage-based unmanned aerial vehicle identification method

Abstract

The invention relates to an unmanned aerial vehicle identification method based on linkage of a decision tree and a radial basis SVM (Support Vector Machine), which can be used for a narrow-band radar to classify a detected target and judge whether the detected target is an unmanned aerial vehicle or not. The method comprises the following steps: acquiring a frequency spectrum image of a target, then extracting five types of features including a target speed, a micro-motion feature, a radar cross section (RCS), a difference value between a main peak and a secondary peak, and a difference value between the main peak of the target and left and right points thereof to form a feature vector, and inputting the feature vector into a decision tree to carry out threshold judgment step by step. And stopping the identification process and outputting an identification result if the target can be judged and identified at any level, continuing identification if the target cannot be judged and identified, and inputting the characteristics of the target which cannot be identified after the decision tree runs into the radial basis SVM to complete identification of the unmanned aerial vehicle. According to the invention, the problem that high-precision identification of long and short distance unmanned aerial vehicles is difficult at the same time under the condition that the narrow-band radar cannot acquire the target height is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of target recognition, and in particular relates to a method for identifying unmanned aerial vehicles (UAVs) based on the linkage of a decision tree and a radial basis support vector machine (SVM), which is used for identifying a target detected by a narrow-band radar and determining whether the target is a UAV. Background Art

[0002] Drones are typical "low, slow, and small" targets. With the rapid development of autonomous networking technology, dynamic task adjustment technology, and payload miniaturization technology, drones have greatly enhanced their situational awareness and anti-interference capabilities, and have the advantages of high flexibility, high maneuverability, and strong survivability. In addition to their simple and easy operation, small size, good concealment, and low cost, drones are gradually being used for aerial photography and entertainment in addition to agriculture, environmental monitoring, and other industries. As the scope of use gradually expands, illegal drone flights have become a new focus, posing severe challenges to air route safety and urban security. Therefore, the need to detect and identify drones has also emerged.

[0003] Analyzing the Doppler effect and micro-Doppler effect of echo signals is of great significance for detecting and identifying drones using radar technology.

[0004] The Doppler effect is a phenomenon in which the main frequency band of the echo is offset due to the movement of the main body of the radar detection target. The micro-Doppler effect refers to the additional modulation phenomenon of the target echo outside the main Doppler frequency band due to the micro-movement of the target such as vibration and rotation in addition to the translational Doppler frequency of the main body of the radar detection target. When the target structure is different, there are certain differences in the micro-motion phenomenon caused by the movement, and the main Doppler is closely related to factors such as the material and main movement of the target. Therefore, by extracting features from the main Doppler and micro-Doppler modulation parameters of the target echo signal, the classification and recognition of the target can be achieved.

[0005] In recent years, the use of Doppler effect and micro-Doppler effect to identify drones has aroused extensive discussion among radar scholars at home and abroad, and some cutting-edge methods have been proposed. In 2016, Hoffinann F of Germany proposed a method for detecting and tracking drones using a multi-base radar network. The method of combining time domain and micro-Doppler characteristics was used to detect drone echoes, which improved the resolution of ground clutter and drone echoes. However, when a large number of targets appear in the air, the use of micro-Doppler feature analysis and high-resolution imaging technology will require a long target time interval and cannot identify targets well; in 2018, Wang Yunpeng et al. proposed using smoothed pseudo-Wigner-Ville distribution (Smoothed PWVD, SPWV) to perform time-frequency analysis on simulated drone detection echoes, which can calculate the typical characteristic parameters of different drone rotor shapes, but because the experiment is built based on theoretical simulation, the measured results need further analysis. It can be seen from the text descriptions of the two methods that both use time domain or time-frequency analysis methods, which are also the commonly used analysis methods for identifying drones using the Doppler effect and micro-Doppler effect. This results in the drone being unable to be identified at the same time as it is detected, and it must undergo a certain period of signal analysis before it can be judged, so the real-time performance is not optimal; in addition, in order to obtain more Doppler features, many methods also require that the radar must have a sufficiently high resolution, and there are rigid requirements for the bandwidth of the radar signal, which results in the current research on drone identification methods based on low-resolution radar, i.e., narrow-band radar. There is still a gap.

[0006] The target echo signal of the low-resolution radar also contains the main Doppler characteristics and micro-Doppler characteristics of the target. The main Doppler and micro-Doppler modulation parameters are designed and features are extracted. Some other targets (hereinafter referred to as noise) are eliminated through the decision tree, and some drones are picked out. The drones and noise that cannot be distinguished are classified and identified through the radial basis support vector machine (SVM), which can realize Doppler drone identification based on the linkage of decision tree and radial basis SVM. Summary of the invention

[0007] The technical problems to be solved by the present invention are:

[0008] In order to avoid the shortcomings of the prior art, the present invention provides a UAV identification method based on the linkage of a decision tree and a radial basis support vector machine, which is used to solve the problem that it is difficult to simultaneously perform high-precision identification of UAVs at long and short distances when a narrow-band radar cannot obtain the target altitude.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0010] A method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine, characterized by comprising:

[0011] Acquire the spectrum image of the target;

[0012] Extract five types of features based on the spectrum image, the five types of features include: target speed, micro-motion characteristics, radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the main peak of the target and its left and right points; form the five types of features into a feature vector;

[0013] The feature vector is input into the decision tree for threshold judgment at each level. If the target can be identified at any level, the recognition process can be stopped and the recognition result can be output. If it cannot be identified, the recognition can be continued. If the target cannot be identified after the decision tree is run, the feature vector is input into the radial basis SVM;

[0014] The final recognition of the drone is completed through radial basis support vector machine.

[0015] A further technical solution of the present invention: the formula for determining the target speed is as follows:

[0016]

[0017] Among them, f d is the main Doppler frequency deviation, v is the radial velocity of the target relative to the radar, and λ is the wavelength of the radar transmitted signal.

[0018] A further technical solution of the present invention: the formula for determining the micro-motion feature is as follows:

[0019]

[0020] Where n is the number of Doppler units contained in the micro-Doppler modulation area, [a1, a2, ..., a n ] is the amplitude vector of each Doppler unit.

[0021] A further technical solution of the present invention: the formula for determining the radar cross-section is as follows:

[0022]

[0023] in, and P r It is the radar received power measured when the calibration object with known precise RCS and the target to be measured are placed at the same distance under the condition that other conditions are the same.

[0024] A further technical solution of the present invention is: the step of inputting the feature vector into the decision tree to perform threshold judgment step by step includes:

[0025] Compare the target speed with the maximum speed of the drone. If the target speed exceeds the maximum speed, it is judged as a noise point, otherwise it continues to move downward;

[0026] Compare the micro-motion feature of the target with the micro-motion threshold to determine whether the micro-motion feature of the target is obvious; if the micro-motion feature is obvious, enter the first branch; if the micro-motion feature is not obvious, enter the second branch;

[0027] The first branch identifies drones based on radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points.

[0028] The second branch identifies drones based on radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points.

[0029] A further technical solution of the present invention: The method for determining the micro-motion threshold is to obtain it by multiplying the system noise by a proportional factor.

[0030] A further technical solution of the present invention is as follows: the first branch sequentially performs the steps of identifying the drone based on the radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points, including:

[0031] Determine whether the radar cross-sectional area of ​​the target is within the first threshold range, if yes, continue to move downward, otherwise, determine it as a noise point and stop identification;

[0032] Determine whether the difference between the main peak and the secondary peak is greater than the first threshold value. If yes, continue to move downwards; otherwise, determine it as a noise point and stop recognition;

[0033] It is determined whether the difference between the main peak of the target and its left and right points is within the second threshold range. If so, the target is determined to be a drone, otherwise it is determined to be a noise point.

[0034] A further technical solution of the present invention is that the first threshold value is negatively correlated with the radial distance of the target to the radar, and the values ​​of a and b can be determined by fitting the following linear equation based on empirical values:

[0035] ΔA a1 =-ax+b

[0036] Here, x represents the distance.

[0037] A further technical solution of the present invention: The second branch sequentially performs the steps of identifying the drone based on the radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points, including:

[0038] Determine whether the RCS feature of the target is within the range of the third threshold. If yes, continue to move downward. Otherwise, it is determined as a noise point and stops recognition.

[0039] Determine whether the difference between the main peak and the secondary peak is greater than the second threshold value ΔA b1 If yes, continue to move down, otherwise it is judged as a noise point and stop recognition;

[0040] Determine the difference T between the target main peak and its left and right points dif Whether it is within the range of the fourth threshold, if yes, continue to move downward, otherwise it is determined as a noise point and stops recognition;

[0041] It is determined whether the difference between the main peak and the secondary peak is greater than the third threshold value. If so, the target is determined to be a drone. Otherwise, the feature vector of the target is retained for subsequent input into the radial basis support vector machine for identification.

[0042] A computer system, characterized in that it comprises: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0043] The beneficial effects of the present invention are:

[0044] The present invention provides a method for identifying drones based on the linkage of a decision tree and a radial basis support vector machine, which extracts multiple types of features of the main Doppler and micro-Doppler of the target echo, and inputs the features into a decision tree to distinguish whether it is a drone or a noise point. The radial basis support vector machine is used to make a final judgment on the targets that cannot be distinguished, thereby realizing the identification of drones. This method solves the problem that it is difficult to simultaneously perform high-precision identification of drones at long and short distances when the narrow-band radar cannot obtain the target height. It has the following advantages:

[0045] 1) The present invention has excellent real-time performance. It can be applied in engineering on narrow-band radars and can identify a single-frame target echo signal. It has short accumulation time and processing time, and can detect and complete the identification task at the same time as the target.

[0046] 2) It is suitable for low signal-to-noise ratio environments and can identify both drones with high signal-to-noise ratio and obvious micro-motion characteristics and drones with low signal-to-noise ratio and no obvious micro-motion characteristics;

[0047] 3) High recognition accuracy and low false alarm rate, which can provide relatively reliable information for the subsequent UAV countermeasures;

[0048] 4) It can be applied to low-resolution radar, i.e. narrow-band radar, and the engineering implementation is easy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings are only for the purpose of illustrating specific embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0050] Figure 1 Flowchart of the implementation steps of the drone identification method based on the linkage of decision tree and radial basis support vector machine.

[0051] Figure 2Spectrum waveform of a drone (DJI Phantom 4).

[0052] Figure 3 The positions of the main and secondary peaks in the spectrum waveform of a drone (DJI Phantom 4).

[0053] Figure 4 Flowchart of the decision tree for drone identification.

[0054] Figure 5 The third group of screenshots of the drone’s dots displayed on the terminal during flight. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0056] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0057] The present invention provides a method for identifying unmanned aerial vehicles based on the linkage of a decision tree and a radial basis support vector machine. First, a frequency spectrum image of a target is acquired. Then, five types of features, including target speed, micro-motion feature, radar cross-section (RCS), difference between a main peak and a secondary peak, and difference between a main peak of the target and two points on its left and right, are extracted to form a feature vector. The feature vector is input into a decision tree for threshold judgment step by step. If the target can be identified at any level, the identification process can be stopped and the identification result can be output. If not, the identification can be continued. For the target that cannot be identified after the decision tree is run, its features are input into a radial basis support vector machine to complete the final identification of the unmanned aerial vehicle.

[0058] like Figure 1 As shown, the following steps are included:

[0059] (1) Obtain the spectrum waveform of the target;

[0060] (2) Extract five types of features from the target’s spectral waveform to form a feature vector;

[0061] (3) Input the feature vector into the decision tree and perform threshold judgment step by step;

[0062] (4) For targets that cannot be identified by the decision tree, their feature vectors are input into the radial basis support vector machine for drone identification.

[0063] In order to enable those skilled in the art to better understand the present invention, the present invention is described in detail below in conjunction with specific embodiments.

[0064] (1) Obtain the target’s spectrum waveform.

[0065] The target's spectral waveform includes the main Doppler modulation and micro-Doppler modulation caused by the target's main motion. These two types of Doppler modulation contain rich target features such as the target's shape, material, and movement speed. This information can be used for target recognition. Figure 2 This is the spectrum waveform of the drone (DJI Phantom 4), where you can see obvious main Doppler modulation and micro Doppler modulation.

[0066] (2) Extract five types of features from the target’s spectral waveform.

[0067] According to the Doppler modulation of the target spectrum waveform, five types of features are designed, including target speed, micro-motion characteristics, radar cross-section (RCS), the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points. The features extracted from each spectrum waveform form a feature vector containing 5 feature elements.

[0068] a) Speed ​​characteristic T v

[0069] The cruising speed of a small multi-rotor drone is about 30-50 km / h, and the maximum speed is about 70 km / h. From the Doppler effect, we can see that the main Doppler frequency deviation is positively correlated with the target's moving speed, which can be expressed by formula (1):

[0070]

[0071] Among them, f d is the main Doppler frequency deviation, v is the radial velocity of the target relative to the radar, and λ is the wavelength of the radar transmit signal. According to the above formula, the radial velocity of the target relative to the radar can be calculated.

[0072] b) Micro-motion feature T md

[0073] The vibration, rotation and other micro-motions of the target will cause the target echo to have additional modulation outside the main Doppler band, which is called micro-Doppler modulation. Figure 2The rotation speed of the top of the drone blade can reach 100-150m / s. The micro-Doppler modulation can occupy a considerable bandwidth. The modulation area contains rich target information. Therefore, the amplitude of the micro-Doppler modulation area can be used as the micro-motion feature of the target. The micro-motion feature designed by the present invention is to calculate the average value of the micro-Doppler modulation area. Assuming that the number of Doppler units contained in the micro-Doppler modulation area is n, the amplitude of each Doppler unit can be expressed as a vector [a1, a2, ..., a n ], micro motion characteristic T md The calculation can be expressed by formula (2):

[0074]

[0075] c) Radar Cross Section (RCS) T r

[0076] It refers to the echo strength of the reflected signal of a radar signal irradiating a single target object after being received by the receiver. RCS is determined by the characteristics of the target itself and has nothing to do with the characteristics of the radar signal and factors such as distance. It is an important feature for radar target classification. The target RCS can be measured using the relative calibration method and can be expressed using formula (3):

[0077]

[0078] In formula (3) and P r It is the radar received power measured when the calibration object with known precise RCS and the target to be measured are placed at the same distance under the condition that other conditions are the same.

[0079] d) The difference between the main peak and the secondary peak T Δ

[0080] The main peak reflects the strength of the target's reflection of the radar signal, including information such as the target's shape, material, and size. The secondary peak contains the target's micro-motion information. Therefore, the difference between the main peak and the secondary peak can be used to distinguish between drones and noise. The threshold of this eigenvalue is related to the distance, and a fitting equation is needed to determine the threshold for different distances. Figure 3 The main peak and secondary peak positions in the spectrum waveform of the drone (DJI Phantom 4) are marked.

[0081] e) The difference between the target main peak and its left and right points T dif

[0082] The radar main peak contains information such as the shape, material, and size of the target. Therefore, in addition to the peak value of the main peak, the difference between the main peak and its left and right points can also be extracted as a feature to distinguish between drones and noise.

[0083] (3) Input the features into the decision tree and perform threshold judgment step by step.

[0084] Figure 4 This is a flowchart for drone identification using a decision tree. The feature vector is input into the decision tree from the top, and the decision tree begins to perform threshold judgments at each level. If the target type can be determined at any level, it stops and outputs the recognition result. If not, it continues to identify. For targets that cannot be identified after the decision tree is finished running, its feature vector is input into the radial basis SVM to further complete the identification of the drone.

[0085] Detailed process of decision tree recognition:

[0086] 1) Input the feature vector into the decision tree;

[0087] 2) The target's velocity characteristic T v The maximum speed of the drone v max For comparison, if the target speed exceeds v max , it is judged as a miscellaneous point, otherwise it continues downward;

[0088] 3) The micro-motion feature T of the target md Compared with the micro-motion threshold, it is judged whether the micro-motion feature of the target is obvious. Here, the decision tree is divided into two branches. The micro-motion threshold is obtained by multiplying the system noise by a proportional factor. In the figure, n is the system noise and η is the self-set proportional factor.

[0089] 4) Branch A considers that the target has obvious micro-motion characteristics.

[0090] 4-1) After entering the A branch, first determine whether the target's RCS characteristics belong to (R a1 ,R a2 ) range, if yes, continue to go down to 4-2), otherwise it is judged as a noise point and stop recognition;

[0091] 4-2) Determine the difference T between the main peak and the secondary peak Δ Is it greater than the threshold value ΔA? a1 If yes, continue to go down to 4-3), otherwise it is judged as noise and stop recognition. Threshold value ΔA a1 The value of is negatively correlated with the radial distance of the target to the radar. The values ​​of a and b can be determined by fitting the linear formula (4) based on the empirical value. x represents the distance

[0092] ΔA a1 =-ax+b (4)

[0093] 4-3) Determine the difference T between the target main peak and its left and right points dif Does it belong to (D a1 ,D a2) range, if it is, the target is determined to be a drone, otherwise it is determined to be a noise point, and the A branch completes the identification.

[0094] 5) Branch B considers that the micro-motion characteristics of the target are not obvious.

[0095] 5-1) After entering the B branch, first determine whether the target's RCS characteristics belong to (R b1 ,R b2 ) range, if yes, continue to go down to 5-2), otherwise it is judged as a noise point and stop recognition;

[0096] 5-2) Determine the difference T between the main peak and the secondary peak Δ Is it greater than the threshold value ΔA? b1 If yes, continue to go down to 5-3), otherwise it is judged as noise and stop recognition. Threshold value ΔA b1 The value is an experience value.

[0097] 5-3) Determine the difference T between the target main peak and its left and right points dif Does it belong to (D b1 ,D b2 ) range, if yes, continue to go down to 5-4), otherwise it is judged as a noise point and stop recognition;

[0098] 5-4) Determine the difference T between the main peak and the secondary peak Δ Is it greater than the threshold value ΔA? b2 (Threshold value ΔA b2 The value of is an empirical value). If it is, the target is determined to be a drone. Otherwise, the feature vector of the target is retained for subsequent input into the radial basis support vector machine for identification.

[0099] This concludes the decision tree's recognition of drones.

[0100] (4) For targets that cannot be identified by the decision tree, their feature vectors are input into the radial basis support vector machine for drone identification.

[0101] The support vector machine used in this method is a radial basis binary classification vector machine. The feature vector of the target that cannot be identified in step (3) is input into the support vector machine. The output result is 1, indicating that the target is a drone, and the output result is -1, indicating that the target is a noise point.

[0102] In order to further evaluate the superiority of the present invention in drone target recognition, the method proposed in the present invention is implemented on a certain type of radar, that is, an engineering implementation is carried out. The radar is a narrow-band radar and cannot accurately measure the pitch angle and height information of the target for drone recognition. Three groups of drone flight tests are carried out at three different distances. The target point trace recognition results are output together with the point trace information. Different targets are distinguished by different colors in the terminal. After that, the recognition rate of the drone and the false alarm rate of the noise are respectively counted. The statistical data are listed in Table 1. Figure 5 This is a screenshot of the third group of drones showing dots on the terminal during flight. Red indicates they are identified as drones, and yellow indicates they are identified as noise dots.

[0103] Table 1 Statistics of recognition rate of drones at different distances and false alarm rate of clutter

[0104]

[0105] By analyzing the data in Table 1, we find that:

[0106] 1. The method proposed in the present invention has a high signal-to-noise ratio for UAVs at close distances, and the recognition rate can reach more than 90%. For UAVs at long distances, the signal-to-noise ratio will decrease significantly with the increase of distance, but the recognition rate can be maintained at about 84%;

[0107] 2. The method proposed in the present invention has a false alarm rate of about 10%.

[0108] observe Figure 5 ,Discover:

[0109] With an identification rate of 84.97%, a false alarm rate of 10.42%, and a low signal-to-noise ratio of 12dB to 27dB, the drone can be clearly distinguished from other targets.

[0110] Through the experimental process and the above analysis, it can be seen that the method proposed in the present invention has excellent real-time performance, and can perform and complete the recognition task while the target point traces are detected; secondly, it can recognize both UAVs with high signal-to-noise ratio in the near area and obvious micro-motion characteristics and UAVs with low signal-to-noise ratio in the far area and unclear micro-motion characteristics, and it has high recognition accuracy and low false alarm rate.

[0111] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should be included in the protection scope of the present invention.

Claims

1. A drone identification method based on decision tree and radial basis support vector machine linkage, characterized in that: include: Acquire the spectrum image of the target; Extract five types of features based on the spectrum image, the five types of features include: target speed, micro-motion characteristics, radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the main peak of the target and its left and right points; form the five types of features into a feature vector; The feature vector is input into the decision tree for threshold judgment at each level. If the target can be identified at any level, the recognition process can be stopped and the recognition result can be output. If it cannot be identified, the recognition can be continued. If the target cannot be identified after the decision tree is run, the feature vector is input into the radial basis SVM; The final recognition of the drone is completed through radial basis support vector machine.

2. According to claim 1, a method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine, characterized in that: The target speed is determined by the following formula: Among them, f d is the main Doppler frequency deviation, v is the radial velocity of the target relative to the radar, and λ is the wavelength of the radar transmitted signal.

3. The method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine according to claim 1, characterized in that: The formula for determining the micro-motion characteristics is as follows: Where n is the number of Doppler units contained in the micro-Doppler modulation area, [a1, a2, ..., a n ] is the amplitude vector of each Doppler unit.

4. The method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine according to claim 1, characterized in that: The formula for determining the radar cross section is as follows: in, and P r It is the radar received power measured when the calibration object with known precise RCS and the target to be measured are placed at the same distance under the condition that other conditions are the same.

5. The method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine according to claim 1, characterized in that: The step of inputting the feature vector into a decision tree to perform threshold judgment step by step includes: Compare the target speed with the maximum speed of the drone. If the target speed exceeds the maximum speed, it is judged as a noise point, otherwise it continues to move downward; Compare the micro-motion feature of the target with the micro-motion threshold to determine whether the micro-motion feature of the target is obvious; if the micro-motion feature is obvious, enter the first branch; if the micro-motion feature is not obvious, enter the second branch; The first branch identifies drones based on radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points. The second branch identifies drones based on radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points.

6. The method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine according to claim 5, characterized in that: The method for determining the micro-motion threshold is to obtain it by multiplying the system noise by a proportional factor.

7. The method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine according to claim 5, characterized in that: The first branch sequentially performs the steps of identifying the drone based on the radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points, including: Determine whether the radar cross-sectional area of ​​the target is within the first threshold range, if yes, continue to move downward, otherwise, determine it as a noise point and stop identification; Determine whether the difference between the main peak and the secondary peak is greater than the first threshold value. If yes, continue to move downwards; otherwise, determine it as a noise point and stop recognition; It is determined whether the difference between the main peak of the target and its left and right points is within the second threshold range. If so, the target is determined to be a drone, otherwise it is determined to be a noise point.

8. The method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine according to claim 7, characterized in that: The first threshold value is negatively correlated with the radial distance of the target to the radar. The values ​​of a and b can be determined by fitting the following linear equation based on empirical values: ΔA a1 -ax+b Here, x represents the distance.

9. The method for identifying unmanned aerial vehicles based on the linkage of decision tree and radial basis support vector machine according to claim 5, characterized in that: The second branch sequentially performs the steps of identifying the drone based on the radar cross-sectional area, the difference between the main peak and the secondary peak, and the difference between the target main peak and its left and right points, including: Determine whether the RCS feature of the target is within the range of the third threshold. If yes, continue to move downward. Otherwise, it is determined as a noise point and stops recognition. Determine whether the difference between the main peak and the secondary peak is greater than the second threshold value ΔA b1 If yes, continue to move down, otherwise it is judged as a noise point and stop recognition; Determine the difference T between the target main peak and its left and right points dif Whether it is within the range of the fourth threshold, if yes, continue to move downward, otherwise it is determined as a noise point and stops recognition; It is determined whether the difference between the main peak and the secondary peak is greater than the third threshold value. If so, the target is determined to be a drone. Otherwise, the feature vector of the target is retained for subsequent input into the radial basis support vector machine for identification.

10. A computer system, characterized in that include: One or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method of claim 1.

Images