DE-SVM-based online evaluation method for countering interference effect of small civil unmanned aerial vehicle

Abstract

The embodiment of the invention discloses a DE-SVM-based online evaluation method for a countering interference effect of a small civil unmanned aerial vehicle, and belongs to the technical field of unmanned aerial vehicle countering. According to the method, when it is detected that a non-cooperative target enters a prevention and control area, motion characteristic parameters of the target are continuously extracted; after the interference is performed for a preset time, identifying the flight state of the target by adopting an identification model based on a support vector machine based on the motion characteristic parameters of the target; and evaluating a countering interference effect based on the flight state of the target. According to the method, the target interference effect evaluation model of the target flight state and the target motion characteristic quantity and the problem solving model of the support vector machine combined with action decomposition are constructed, so that the accuracy of interference effect identification based on the support vector machine is improved, and online and cognition of the target interference effect evaluation method are further realized.

Description

Technical Field

[0001] The invention belongs to the technical field of unmanned aerial vehicle countermeasures, and relates to an online evaluation method for the countermeasure interference effect of a small civilian unmanned aerial vehicle based on DE-SVM. Background Art

[0002] As the venue for large-scale events, once large venues are disturbed by drones, it will not only affect the event process and event arrangements, but also cause great harm to the people participating in the event, easily causing crowd panic and stampede, and posing a huge safety hazard. At the same time, large venues are generally located in densely populated urban areas. In order to prevent secondary damage to the ground, soft-kill measures are mainly used to interfere with and counter drones.

[0003] Soft kill mainly implements directional suppression interference on the navigation signal and remote control information of the drone, or GPS navigation interference, forcing the drone to automatically return, automatically descend, or hover. For interference effect evaluation, most of them are to judge the posture of the threat target through radar after the interference is implemented, so as to judge whether the threat target is directly falling, automatically returning, automatically descending, or hovering in the air. Figure 1 An online evaluation method for interference effects based on the combination of pulse vision and statistical features is shown.

[0004] The traditional counter-interference effect evaluation is to obtain the effect of the interference with the help of human subjective judgment after the interference is implemented. With the increasing anti-interference ability and intelligence level of drones, this method can no longer adapt to the current complex and changeable confrontation scenarios. Summary of the invention

[0005] The purpose of the present invention is to provide an online evaluation method for the counter-interference effect of a small civilian UAV based on DE-SVM, so as to improve the accuracy of interference effect identification and realize the online and cognitive evaluation method of the target interference effect.

[0006] A first aspect of the present invention provides an online evaluation method for the counter-interference effect of a UAV, comprising:

[0007] When a non-cooperative target is detected entering the control area, the target's motion characteristic parameters are continuously extracted; wherein the motion characteristic parameters include the route shortcut change ΔD, the route shortcut standard deviation σ D , height change Δh, height standard deviation σ h , standard deviation of height variation σ Δh And the heading angle change Δθ;

[0008] After the predetermined interference time, the flight state of the target is identified based on the motion characteristic parameters of the target using a recognition model based on a support vector machine; wherein in the recognition model, the multi-classifier of the support vector machine is composed of two sub-classifiers; the flight state includes returning, hovering, falling and descending;

[0009] evaluating the counter-jamming effect based on the flight state of the target;

[0010] Wherein, using a recognition model to identify the flight state of the target based on the motion characteristic parameters of the target includes:

[0011] The predetermined time is divided into a plurality of unit time slices. In each unit time slice, the first subclassifier is used to extract the route shortcut change ΔD and the route shortcut standard deviation σ from the motion feature parameters reflecting the horizontal motion state of the target. D and the heading angle change Δθ to identify the target's flight action in the horizontal direction, and the second sub-classifier is used to extract the height change Δh and the height standard deviation σ reflecting the vertical motion state of the target from the motion feature parameters h and the standard deviation of height variation σ Δh , to identify the target's flight movements in the vertical direction;

[0012] The flight actions of multiple unit time slices are spliced ​​to output the flight status of the target within the predetermined time.

[0013] In another example, the counter-interference effect is evaluated based on the flight state of the target, including: if the flight state of the target is falling, outputting a first interference effect value; if the flight state of the target is returning, hovering or descending, outputting a second interference effect value.

[0014] In another example, the evaluation of the anti-interference effect based on the flight state of the target further includes: if the flight state of the target is another state, such as a normal state, determining whether the target has performed an anti-interference action based on the signal parameters of the target.

[0015] In another example, the signal parameter includes a change in the number of frequency hopping points and / or a change in the frequency hopping rate.

[0016] In another example, if the target does not perform the anti-interference action or the anti-interference action does not change, the third interference effect value is output; otherwise, the fourth interference effect value is output.

[0017] In another example, when training the recognition model, a differential evolution algorithm is used to optimize the support vector machine parameters.

[0018] In another example, when optimizing the support vector machine parameters, software simulation is used to simulate the target's return, descent, fall and hovering motion trajectories to establish a flight state library and extract the target track information; each type of flight state track in the flight state library is divided into several segments, and its motion feature parameters are extracted respectively to obtain a motion feature parameter set, and the motion feature parameter set and the corresponding flight state category are constructed as a training sample set.

[0019] In another example, the support vector machine adopts a Gaussian kernel function.

[0020] A second aspect of the present invention provides an online evaluation device for the counter-interference effect of a UAV, characterized in that it includes:

[0021] The parameter extraction unit is configured to: when a non-cooperative target is detected to enter the control area, continuously extract the target's motion characteristic parameters; wherein the motion characteristic parameters include the route shortcut change ΔD, the route shortcut standard deviation σ D , height change Δh, height standard deviation σ h , standard deviation of height variation σ Δh And the heading angle change Δθ;

[0022] The state recognition unit is configured to: after the predetermined interference time, recognize the flight state of the target based on the motion characteristic parameters of the target using a recognition model based on a support vector machine; wherein in the recognition model, the multi-classifier of the support vector machine is composed of two sub-classifiers; and the flight state includes returning, hovering, falling and descending;

[0023] An effect evaluation unit is configured to: evaluate the counter-interference effect based on the flight state of the target;

[0024] Wherein, the state recognition unit comprises:

[0025] The motion recognition module is configured to: in each unit time slice, use the first sub-classifier to extract the route shortcut change ΔD and the route shortcut standard deviation σ from the motion feature parameters that reflect the horizontal motion state of the target D and the heading angle change Δθ to identify the target's flight action in the horizontal direction, and the second sub-classifier is used to extract the height change Δh and the height standard deviation σ reflecting the vertical motion state of the target from the motion feature parameters h and the standard deviation of height variation σ Δh , to identify the target's flight action in the vertical direction; wherein the unit time slice is obtained by dividing the predetermined time;

[0026] The state recognition module is configured to: splice the flight actions of multiple unit time slices to output the flight state of the target within the predetermined time.

[0027] The present invention improves the accuracy of interference effect recognition based on a support vector machine, and can realize the online and cognitive evaluation method of a target interference effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flow chart of a known method for online evaluation of interference effects based on the combination of pulse vision and statistical features;

[0029] Figure 2 Flow chart of the online evaluation method of the UAV counter-interference effect according to the embodiment of the present invention;

[0030] Figure 3 A schematic diagram of a target interference effect evaluation model according to an embodiment of the present invention;

[0031] Figure 4 The following is a schematic diagram of the target flight state action division; (a) is a left turn and return, (b) is a right turn and return, and (c) is a vertical descent;

[0032] Figure 5 A schematic diagram of the principle of a flight status recognition model according to an embodiment of the present invention;

[0033] Figure 6 Flowchart for optimizing support vector machine parameters for differential evolution algorithm. DETAILED DESCRIPTION

[0034] The present invention is described in more detail below with reference to the accompanying drawings:

[0035] The present invention constructs an online evaluation model for interference effect by integrating the flight status, track behavior, and anti-interference behavior indicators of low-altitude non-cooperative targets (for example, small civilian drones), and then evaluates the effect of interference on the threatening target to determine the level of interference effect.

[0036] The jamming effect evaluation process for low-altitude non-cooperative targets includes the following steps:

[0037] (1) When the detection system finds a non-cooperative target in the control area, it extracts the corresponding motion feature parameters and selects a suitable suppression interference method to interfere with the target according to the target threat level. The threat level assessment and the suppression interference method selection can be implemented by using known technologies, and the present invention is not limited to this, so the details are not described in detail.

[0038] (2) After a period of interference, the relevant motion characteristic parameters of the target are processed to determine the flight state of the target; if the target is in a falling state compared with the previous state, it means that the target interference effect is strong, and the corresponding interference effect evaluation value is output; if the target is in one of the automatic descent, return, or hovering states, it means that the target interference effect is relatively strong, and the corresponding interference effect evaluation value is output; otherwise, if the target is in a relatively normal flight state, further judgment is required;

[0039] (3) If the target is in a relatively normal flight state, such as basically maintaining the original state, the target's signal parameters can be used to determine whether the target has performed anti-interference behavior or whether the anti-interference behavior has changed. If the target does not have anti-interference behavior, or the anti-interference behavior remains unchanged, the target interference is considered ineffective. If the target's anti-interference behavior has changed, it means that the target interference effect is poor.

[0040] After being subjected to suppressive interference, the main flight states of the drone include: direct fall, automatic descent, automatic return, hovering in the air, etc.

[0041] (1) Direct fall: After being interfered with, if part or all of the links of the drone are interrupted, the drone will directly fall due to program errors. This usually happens because the drone is subject to strong interference or has low intelligence.

[0042] (2) Automatic descent: When the drone’s data transmission, image transmission, and navigation links are disturbed, it executes the descent command according to the automatic landing process set in its own flight control algorithm.

[0043] (3) Automatic return: The data and image transmission links of the drone are subject to suppressive interference, but the navigation link communication is still working normally. The drone completes the return operation according to the program set in the flight control algorithm that the drone automatically returns after losing control.

[0044] (4) Hovering in the air: After a drone is subjected to suppression interference, if its flight control program is set to continue to stay in the air after being interfered with, the drone will appear to hover in the air. Therefore, the effectiveness of target interference can be directly or indirectly judged based on the change in the flight status of the threatening target.

[0045] The flight status of the target after being disturbed is mainly direct falling, automatic return, automatic landing and hovering. The target flight status is related to the target motion parameters output by the sensor, mainly the change of route shortcut, the standard deviation of route shortcut, the change of altitude, the standard deviation of altitude, the standard deviation of altitude change, and the change of heading angle.

[0046] (1) Target route shortcut change ΔD

[0047] The change in the route shortcut reflects the relative position change information of the target on the horizontal plane relative to the center of the protection area. Set the judgment threshold ΔD min and ΔD max (ΔD min >0, ΔD max >0), if in the unit time slice, |ΔD| < ΔD min , then it is judged that the horizontal position of the target before and after the time has not changed, and the target may be in a falling or hovering state; if ΔD|>ΔD max , it means that the target has moved on the horizontal plane, and ΔD>0 means that the target is moving away from the center, and ΔD>0 means that the target is approaching the center. The change in the target route shortcut is a characteristic of the two flight states of the target falling and returning. The calculation formula for the change in the target route shortcut ΔD in the time slice t0~t1 is:

[0048]

[0049] in, They represent the target route shortcuts at t0 and t1 respectively.

[0050] (2) Standard deviation of target route shortcut σ D

[0051] Standard deviation of route shortcut σ D It reflects the fluctuation degree of the target horizontal position information within the detection time slice. Let the discrimination threshold σ Dmin , if within the time slice, σ D <σ Dmin , indicating that the overall change of the target's route shortcut is not large, and the horizontal position information fluctuates very little, so it can be considered that the target has always stayed at a certain horizontal position. The target route shortcut standard deviation is a characteristic quantity to determine whether the target is in a hovering state. The target route shortcut standard deviation σ in the time slice t0~t1 D The calculation formula is:

[0052]

[0053] Where N is the number of sampling points of the route shortcut in the time slice t0~t1, Represents the mean value of route shortcuts.

[0054] (3) Target height change Δh

[0055] The target height change Δh reflects the position information of the target on the vertical plane relative to the protection center area. Set the discrimination threshold Δh max1 and Δh max2 , 0<Δh max1 <Δh max2 If within the time slice, -Δhmax2 <Δh<-Δh max1 , indicating that the height of the target on the vertical plane is decreasing, but the change is not large, and the target may be in an automatic descending flight state; if Δh<-Δh max2 , indicating that the height of the target on the vertical plane changes greatly, and the height of the target relative to the ground is decreasing, indicating that the target is likely to be in a falling state. The target height change is a characteristic of the target's descent and falling state. The calculation formula for the target's height change Δh in the time slice t0~t1 is:

[0056]

[0057] in, They represent the flight altitude of the target at t0 and t1 respectively.

[0058] (4) Target height standard deviation σ h

[0059] Target height standard deviation σ h It reflects the fluctuation degree of the target vertical position information within the detection time slice. Let the discrimination threshold σ hmin , if within the time slice, σ h <σ hmin , it means that the vertical height value of the target does not change much overall, and the vertical position of the target has always fluctuated very little, so it can be considered that the target has always been at a certain vertical height. The target height standard deviation is a characteristic quantity of whether the target is in a hovering state. The target height standard deviation σ in the time slice t0~t1 is h The calculation formula is:

[0060]

[0061] Among them, K is the number of sampling points of the flight altitude in the time slice t0~t1, Represents the mean flight altitude.

[0062] (5) Standard deviation of target height change σ Δh

[0063] Standard deviation of target height change σ Δh It reflects the fluctuation degree of the target vertical height change in the detection time slice. Let the discrimination threshold σ Δhmax , if within the time slice, σ Δh >σ Δhmax , indicating that the target's height changes greatly and it moves at a non-uniform speed in the vertical plane, σ Δh The larger the value, the more dramatic the change in the target's speed in the vertical plane, and the target may be in a falling state. The standard deviation of the target height change is a characteristic quantity of whether the target is falling or falling. The standard deviation of the target's height change in the time slice t0 to t1 isΔh The calculation formula is:

[0064]

[0065] (6) Target heading angle change Δθ

[0066] In order to distinguish the heading angle changes after 0° and 180°, the counterclockwise angle between the target flight direction and the center of the protection area is defined as the positive heading angle, and the clockwise angle is defined as the negative heading angle. The heading angle change of the target within t0~t1 is shown in the following formula:

[0067]

[0068] in, Represent the heading angle of the target at t0 and t1 respectively.

[0069] The target heading angle change is the characteristic quantity for judging the target return state. max , if within the time slice, |Δθ>θ max , indicating that the target has turned. By combining multiple consecutive time slices, if the target has been in a turning state, the route shortcut change information can be used to indirectly determine whether the target is in a return state.

[0070] After taking anti-interference measures, the threat target will usually cause changes in its communication signals. When this signal is intercepted by the reconnaissance equipment, our data fusion center can indirectly evaluate the interference effect through the implementation of the target's anti-interference measures.

[0071] The communication anti-interference measures taken by the drone after being interfered mainly include: spread spectrum anti-interference technology, adaptive interference suppression technology, source and channel coding technology. Most of the existing consumer drones use frequency hopping communication systems to enhance their anti-interference performance. Generally speaking, the larger the frequency hopping bandwidth, the more frequency points of the frequency hopping, and the faster the frequency hopping rate, the stronger the communication anti-interference performance of the target. The present invention mainly considers the change in the number of frequency hopping points and the change in the frequency hopping rate within a period of time before and after the target is interfered to determine whether it takes anti-interference measures.

[0072] (1) Change in the number of target signal frequency hopping points

[0073] When a drone is flying normally, it usually communicates at one or several fixed frequencies to reduce power loss. When it receives interference, the system will increase its frequency hopping points to improve its anti-interference ability. When the detection and receiving system intercepts the change in the frequency hopping points of the drone signal, it can determine whether it has taken anti-interference measures.

[0074] ΔC=C t1 -Ct0

[0075] Among them, C t0 , C t1 They represent the frequency hopping points of the target signal intercepted at time t0 and time t1 respectively.

[0076] (2) Change in target signal frequency hopping rate

[0077] The frequency hopping rate of the drone frequency hopping system usually varies within a range. When working normally, the frequency hopping rate is usually 0 or communicates at a lower frequency hopping rate to reduce communication loss. When the drone takes anti-interference measures, the frequency hopping rate will vary over a larger range.

[0078] ΔH=H t1 -H t0

[0079] Among them, H t0 , H t1 They represent the frequency hopping points of the target signal intercepted at time t0 and time t1 respectively.

[0080] Through the previous analysis, the target's flight state is composed of one or more basic movements. Therefore, it is necessary to comprehensively consider the target's movement state in multiple continuous time segments to further analyze the target's flight state.

[0081] The flight status of the target's return flight can be analyzed in the horizontal plane. The basic movement of the target in the horizontal plane can be divided into three types: straight flight, left turn and right turn. The target's flight direction is counterclockwise for right turn and clockwise for left turn. The symbol s is defined to represent straight-line movement, r for right turn and l for left turn. The action of the target turning left and turning right is divided as follows: Figure 4 As shown in (a) and (b) of FIG. 1 , it can be seen from the figure that the flight state of the target returning home can be expressed as the target moving straight, then turning left or right continuously, and finally changing to straight line motion, which can be used as an important identification basis for the return state.

[0082] For the two states of the target, the movement in the vertical plane is mainly analyzed. The descending flight state of the target can be composed of two kinds of movements, vertical dive and oblique dive in continuous time slices, and the height change in a single time slice is small; the falling flight state of the target is mainly composed of vertical dive movements in continuous time slices, and the height change of the target in a single time is large, which is an important basis for judging the two flight states of the target. Define the symbol ds as the descending action of the target in the vertical plane, then the descending and falling actions of the target in the vertical plane are divided as follows Figure 4 As shown in (c) in .

[0083] For the target hovering flight state, it is necessary to refer to the target's motion parameter variables in the horizontal plane and the vertical plane at the same time. Since the overall position change of this flight state is not large, its actions will not be divided here.

[0084] Through the above analysis, the flight state of the target cannot be analyzed from the behavior in one time slice alone, but needs to be analyzed and identified by combining the flight actions in multiple consecutive time slices. Therefore, for the problem of identifying the flight state of the target, the basic actions of the target in a single time slice can be identified first, and then the basic actions of the target in multiple consecutive time slices can be classified and analyzed, and finally the flight state of the target can be identified.

[0085] Based on the above analysis, the target's flight status is mainly related to the change of the route shortcut ΔD and the standard deviation of the route shortcut σ D , height change Δh, height standard deviation σ h , standard deviation of height variation σ Δh And the heading angle change Δθ are related to these six parameters. Among them, the route shortcut change ΔD, the route shortcut standard deviation σ D The heading angle change Δθ mainly reflects the horizontal motion state of the target, the height change Δh, the height standard deviation σ h , standard deviation of height variation σ Δh It mainly reflects the vertical movement state of the target.

[0086] The present invention uses a support vector machine multi-classifier to perform action recognition based on the segmentation of the target flight state. In essence, it constructs a mapping relationship between the flight actions in different directions and the motion feature parameters that affect the flight actions in this direction. Based on this mapping relationship, the motion feature parameters in multiple continuous time slices are input to further identify the flight actions in the corresponding time slices, and then the actions in the continuous time slices are synthesized, and then the flight state of the target within a period of time can be determined.

[0087] Based on the above analysis, the present invention divides the target flight state within a period of time into flight actions within four consecutive time slices, transforms the flight state recognition problem into the flight action recognition problem, and establishes the following Figure 5 The target flight state recognition model shown.

[0088] exist Figure 5 In , the input dimension of the flight state recognition model is determined to be 6 and the output dimension is 1. For a certain target track information collected, the motion feature parameters in the i-th time slice of the track are input to identify the target flight action in the time slice, then the corresponding mapping function can be expressed by the following formula:

[0089] y i =f(ΔD i ,σDi ,Δh i ,σ hi ,σ Δhi ,Δθ i )

[0090] Among them, y i refers to the target flight action type in the i-th time slice, {ΔD i ,σ Di ,Δh i ,σ hi ,σ Δhi ,Δθ i} are the 6 feature parameter sets within the time slice.

[0091] In this recognition model, the multi-classifier of the support vector machine is composed of two sub-classifiers, which can recognize the target's flight movements in the horizontal direction and the vertical direction respectively. After inputting the motion feature parameters, SVM classifier 1 extracts the motion parameters in the horizontal direction as input, and then recognizes the flight movements in the horizontal direction; SVM classifier 2 extracts the motion parameters in the vertical direction as input, and then recognizes the flight movements in the vertical direction. After obtaining the flight movements of the target in each direction in each divided time slice, each flight movement is spliced ​​and synthesized, thereby outputting the target's flight status within a period of time.

[0092] The model converts the complex flight state classification problem into the equivalent of motion recognition problem in the horizontal and vertical directions through track segmentation and action decomposition. As long as the motion feature parameters are input into the model, the flight state category can be output, thus achieving parameterization and simplification of complex problems.

[0093] In one example, the present invention uses a differential evolution algorithm to optimize the parameters of a support vector machine. The basic calculation steps of the differential evolution algorithm include initializing the population, mutation, crossover, and selection. A new population is generated through continuous evolution, and the global optimal solution is found in the real number space to minimize the fitness value. This section takes the Gaussian kernel support vector machine as an example. The present invention takes minimizing the probability of classification error of the support vector machine as the optimization goal, and optimizes the parameter combination (C, g) composed of the penalty factor C and the kernel parameter g through the differential evolution algorithm. Then the fitness function can be set to the total classification error rate, and the formula is shown below.

[0094]

[0095] Among them, P is the number of test sample labels, Y i is the number of test samples in the ith label, E i is the number of prediction errors in the i-th label.

[0096] The steps for implementing support vector machine parameter optimization based on the difference algorithm are as follows: Figure 6As shown, including:

[0097] (1) Initialize the population

[0098] Define the population size NP, and randomly generate a 2D initial population (C, g) of NP group. Then, in the G-th generation evolution, the composition of the i-th individual can be expressed as:

[0099]

[0100] The jth parameter value of the i-th individual is:

[0101]

[0102] in, The upper bound of the individual vector, Lower bound for the individual vector.

[0103] (2) Mutation Operation

[0104] In the differential evolution algorithm, the basic principle of individual vector mutation is to randomly select two different individuals in the Gth generation evolution iteration. The difference between the two and add other individuals to get a new variant individual The individual variation formula is as follows:

[0105]

[0106] Where a, b∈{1,…,NP} are any integers, a≠b≠i; F is the variation factor, which is a constant, and its value range is usually [0,2]; It refers to the best individual in the G-th generation iteration.

[0107] (3) Crossover operation

[0108] After the mutation task is completed, the newly generated mutant individuals and the individuals to be mutated are cross-mixed to generate new experimental individuals. Commonly used crossover schemes include binomial crossover and exponential crossover. Among them, according to existing studies, the binomial crossover scheme is significantly better than the exponential crossover scheme. Its crossover formula is as follows:

[0109]

[0110] Among them, CR∈[0,1] is the crossover probability constant, j rand ∈{1,2}, is a randomly generated positive integer, j=j rand This constraint ensures that at least one dimension of the mutation vector is replaced in the cross-task.

[0111] (4) Select an operation

[0112] The differential evolution algorithm uses a greedy operator to select individuals from the offspring population. By comparing and calculating the fitness function values ​​corresponding to the current population individuals and the crossover individuals, individuals with better fitness function values ​​are selected as offspring individuals. The calculation formula of the greedy operator is:

[0113]

[0114] In one example, simulation software such as STK is used to simulate the real low-altitude target return, descent, fall and hovering motion trajectories, establish a flight state library for these four flight states, and extract the target track information. Each type of flight state track in the flight state library is divided into four segments and its motion feature parameters are extracted respectively to obtain the motion feature parameter set X = {ΔD,σ D ,Δh,σ h ,σ Δh ,Δθ} as input and the corresponding flight state category as output (label) to construct a training sample set.

[0115] For example, the sample size of each type of flight status data is divided as shown in the following table:

[0116]

[0117]

[0118] In order to avoid the error of classification results caused by the difference in the value range of the input feature parameters, it is necessary to normalize the training sample set and the test sample set. The normalization calculation formula using the range method is as follows:

[0119]

[0120] Among them, X refers to a training feature parameter in the sample set, X max and X min are the maximum and minimum values ​​of the training parameter in the total samples.

[0121] According to another embodiment of the present invention, there is also provided an online evaluation device for the counter-interference effect of a UAV, comprising:

[0122] The parameter extraction unit is configured to: when a non-cooperative target is detected to enter the control area, continuously extract the target's motion characteristic parameters; wherein the motion characteristic parameters include the route shortcut change ΔD, the route shortcut standard deviation σ D , height change Δh, height standard deviation σ h , standard deviation of height variation σ Δh And the heading angle change Δθ;

[0123] The state recognition unit is configured to: after the predetermined interference time, recognize the flight state of the target based on the motion characteristic parameters of the target using a recognition model based on a support vector machine; wherein in the recognition model, the multi-classifier of the support vector machine is composed of two sub-classifiers; and the flight state includes returning, hovering, falling and descending;

[0124] An effect evaluation unit is configured to: evaluate the counter-interference effect based on the flight state of the target;

[0125] Wherein, the state recognition unit comprises:

[0126] The motion recognition module is configured to: in each unit time slice, use the first sub-classifier to extract the route shortcut change ΔD and the route shortcut standard deviation σ from the motion feature parameters that reflect the horizontal motion state of the target D and the heading angle change Δθ to identify the target's flight action in the horizontal direction, and the second sub-classifier is used to extract the height change Δh and the height standard deviation σ reflecting the vertical motion state of the target from the motion feature parameters h and the standard deviation of height variation σ Δh , to identify the target's flight action in the vertical direction; wherein the unit time slice is obtained by dividing the predetermined time;

[0127] The state recognition module is configured to: splice the flight actions of multiple unit time slices to output the flight state of the target within the predetermined time.

[0128] Example:

[0129] In this embodiment, a support vector machine is used to select four different kernel functions, set the kernel parameter g=100, and the penalty factor C=1. The model training is performed 10 times on the training samples in the above table, and then the test samples are recognized. The average of the recognition accuracy and the running time of 10 times are output and compared as shown in the following table:

[0130] Kernel Function Linear Kernel Sigmoid kernel Gaussian Kernel Polynomial Kernel Average recognition accuracy (%) 85.91％ 78.22％ 92.86％ 89.74％ Run time(s) 7.36 9.35 3.05 13.6

[0131] It can be seen that compared with the other three kernel functions, the average recognition accuracy calculated by selecting the Gaussian kernel function is higher and the calculation time is faster.

[0132] In addition, by comparing the predicted types and actual types of the test samples obtained by using the above four functions for sample training, it can be seen that the linear kernel function has 12 errors in 90 predicted samples, the Sigmoid kernel function has 16 errors, the Gaussian kernel function has 6 errors, and the polynomial kernel function has 8 errors. In summary, in processing nonlinear classification recognition models, the Gaussian kernel function has higher recognition accuracy and speed than the kernel function.

[0133] The kernel parameter g and penalty factor C in the Gaussian kernel function are two important parameters. The size of the parameter value directly affects the accuracy of predicting and identifying the target flight state, but it does not mean that the larger the parameter value, the better the classification effect; the size of the penalty factor value affects the penalty intensity of model classification. The larger the C value, the greater the penalty for model misclassification, and the lower the fault tolerance rate. While improving the classification accuracy, it is also easy to cause model overfitting; the smaller the C value, the higher the fault tolerance rate, which is easy to cause low classification accuracy and model underfitting problems. The Gaussian kernel parameter determines the dimension of the mapping and reflects the complexity of the data distribution. The larger the g value, the stronger the kernel function's ability to divide, the worse the classification effect of fresh samples, and the weaker the generalization ability. When the g value is too small, it will lead to low sample classification accuracy.

[0134] The sample set in the above table is also used for model training sample testing, and the population size is set to 30 and the number of iterations is set to 50. The parameters of the differential evolution optimization algorithm are set as follows: the upper and lower limits of the kernel parameter g and the penalty factor C are set to [100, 100] and [0.001, 0.001], the initial crossover probability CR is set to 0.5, and the mutation factor F is set to 0.8. The optimization performance of the differential optimization parameter algorithm is compared with that of the common particle swarm optimization algorithm and the genetic optimization algorithm. After 30 experiments, according to the optimal fitness value optimization results of each optimization model, it can be seen that the particle swarm parameter optimization algorithm has the fastest convergence speed, but its optimal fitness value is the largest, indicating that its recognition accuracy is the lowest; in contrast, the differential evolution optimization algorithm has the smallest optimal fitness value, and the convergence speed is fast, and the algorithm efficiency is high. The experimental results are shown in the following table:

[0135]

[0136]

[0137] From the table above, we can see that under the same population size and number of iterations, compared with the particle swarm optimization algorithm and the genetic optimization algorithm, the support vector machine algorithm model based on the differential evolution algorithm after optimizing the C and g parameters has higher recognition accuracy, more accurate prediction results, and a recognition accuracy rate of 99.21%. From the perspective of algorithm running time, due to the small number of training samples, the running time of each optimization algorithm is fast, but the support vector machine recognition model based on the differential optimization algorithm has the fastest speed in searching for the optimal parameter combination, which is of great significance for the rapid and accurate recognition of the flight status of low-altitude targets in complex transient confrontation environments.

[0138] For the above training test sample data, after using the differential evolution algorithm to optimize the parameters of the support vector machine recognition model, the obtained recognition prediction results were compared with the actual test samples. It was found that when using the trained model for sample recognition prediction, only one sample was recognized incorrectly, and the recognition accuracy was high.

[0139] In addition, it is found that after the differential algorithm is optimized, the convergence value of the penalty factor is 0.94, and the convergence value of the kernel parameter is 1.83, and the highest recognition accuracy is 99.21%. Because the slack variable is related to the value of the penalty factor and the kernel parameter, and the two parameters C and g are randomly generated by the differential intelligent optimization algorithm and iteratively optimized, but the final recognition accuracy is obtained by inputting a large number of samples for comparison, therefore, the optimal parameters of C and g are allowed to fluctuate within a certain range while the recognition accuracy remains unchanged.

[0140] In the STK simulation scenario, the flight status of the target before and after the interference is simulated. Assuming that when 6 low-altitude aircraft are detected invading the low-altitude control area at the same time, the relevant parameters of the invading target are extracted, and the threat index of the invading target is judged to be: 0.82, 0.65, 0.68, 0.60, 0.71, 0.42. After selecting the appropriate interference method and interference parameters, the relevant parameters of the interference threat target are extracted, and the interference effect is evaluated. When calculating the target motion feature quantity and anti-interference index, the sampling parameters with an interval of 5s before and after are taken, and the specific threshold value is set as: ΔD min =5m, ΔD max =10m,σ Dmin =1KH, Δh max1 =10m, Δh max2 =30m,σ hmin =1,σ Δhmax =50,θ max =5°, ΔC max =3, ΔH max =20Hop / s. The possible flight states and anti-interference parameter changes of the threat target in a continuous time period and the corresponding interference effect evaluation results are listed in the following table.

[0141]

[0142]

[0143] The above table shows the target signal parameters collected within a period of time after interference was implemented on 6 threat targets, and the flight status was identified and judged. The flight status of each target was obtained, and the target interference effect evaluation results were obtained by combining the target anti-interference index parameters. It can be seen that after the interference was implemented on target 1, its flight performance was not greatly affected and it was in a normal flight state. It was detected that it did not take anti-interference measures, so the interference on target 1 was invalid, and it was necessary to increase the interference on target 1; the system model identified the flight status of target 2, target 3, and target 5 as return, descent, and hover, respectively, and directly judged that the interference effect on these three targets was strong, and continued to maintain the interference mode; it was identified that target 4 was in a falling state, and it was directly judged that the interference effect on target 4 was strong, and it was necessary to stop the interference on target 4 and take measures to avoid secondary damage caused by the target falling to the ground; it was judged that target 6 was in a normal flight state, but it was detected that the frequency hopping rate of target 6 changed greatly during this period, indicating that target 6 was taking anti-interference measures to counter our interference, so it was judged that the interference effect on target 6 was poor, and the interference intensity needed to be increased.

[0144] For the confrontation scenario between the low-altitude defense system and the invading low-altitude target in a non-combat background, the target interference effect can be evaluated based on the real-time collected target signal parameters and the identified flight status. If the interference intensity is strong enough, continue to interfere with the target in this way or stop the interference; if the interference is ineffective or the effect is poor, it is necessary to improve the interference method according to the actual situation until the target interference effect evaluation result reaches the decision maker's purpose, and then the interference is judged to be successful.

[0145] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the technical solutions described in the above embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents, and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An online evaluation method for the counter-jamming effect of a UAV, characterized in that: include: When a non-cooperative target is detected entering the control area, the target's motion characteristic parameters are continuously extracted; wherein the motion characteristic parameters include the route shortcut change ΔD, the route shortcut standard deviation σ D , height change Δh, height standard deviation σ h , standard deviation of height variation σ Δh And the heading angle change Δθ; After the predetermined interference time, the flight state of the target is identified based on the motion characteristic parameters of the target using a recognition model based on a support vector machine; wherein in the recognition model, the multi-classifier of the support vector machine is composed of two sub-classifiers; the flight state includes returning, hovering, falling and descending; evaluating the counter-jamming effect based on the flight state of the target; Wherein, the flight state of the target is identified by using a recognition model based on the motion characteristic parameters of the target, including: The predetermined time is divided into a plurality of unit time slices. In each unit time slice, the first subclassifier is used to extract the route shortcut change ΔD and the route shortcut standard deviation σ from the motion feature parameters reflecting the horizontal motion state of the target. D and the heading angle change Δθ to identify the target's flight action in the horizontal direction, and the second sub-classifier is used to extract the height change Δh and the height standard deviation σ reflecting the vertical motion state of the target from the motion feature parameters h and the standard deviation of height variation σ Δh , to identify the target's flight movements in the vertical direction; The flight actions of multiple unit time slices are spliced ​​to output the flight status of the target within the predetermined time.

2. The online evaluation method for the UAV counter-interference effect according to claim 1 is characterized in that: The counter-interference effect evaluation based on the flight state of the target includes: if the flight state of the target is falling, outputting a first interference effect value; if the flight state of the target is returning, hovering or descending, outputting a second interference effect value.

3. The online evaluation method for the UAV counter-interference effect according to claim 2 is characterized in that: The evaluating of the anti-interference effect based on the flight state of the target also includes: if the flight state of the target is other states, judging whether the target has performed an anti-interference action according to the signal parameters of the target.

4. The online evaluation method for the UAV counter-interference effect according to claim 3 is characterized in that: The signal parameters include a change in the number of frequency hopping points and / or a change in the frequency hopping rate.

5. The online evaluation method for the UAV counter-interference effect according to claim 3 or 4 is characterized in that: If the target does not perform the anti-interference action or the anti-interference action does not change, the third interference effect value is output; otherwise, the fourth interference effect value is output.

6. The online evaluation method for the UAV counter-interference effect according to claim 1 is characterized in that: When training the recognition model, the differential evolution algorithm is used to optimize the support vector machine parameters.

7. The online evaluation method for the UAV counter-interference effect according to claim 6 is characterized in that: When optimizing the support vector machine parameters, software simulation is used to simulate the target's return, descent, fall and hovering motion trajectories to establish a flight state library and extract the target track information; each type of flight state trajectory in the flight state library is divided into several segments, and its motion feature parameters are extracted respectively to obtain a motion feature parameter set, and the motion feature parameter set and the corresponding flight state category are constructed as a training sample set.

8. The online evaluation method for the UAV counter-interference effect according to claim 6 is characterized in that: The support vector machine uses the Gaussian kernel function.

9. An online evaluation device for the counter-interference effect of a UAV, characterized in that: include: The parameter extraction unit is configured to: when a non-cooperative target is detected to enter the control area, continuously extract the target's motion characteristic parameters; wherein the motion characteristic parameters include the route shortcut change ΔD, the route shortcut standard deviation σ D , height change Δh, height standard deviation σ h , standard deviation of height variation σ Δh And the heading angle change Δθ; The state recognition unit is configured to: after the predetermined interference time, recognize the flight state of the target based on the motion characteristic parameters of the target using a recognition model based on a support vector machine; wherein in the recognition model, the multi-classifier of the support vector machine is composed of two sub-classifiers; and the flight state includes returning, hovering, falling and descending; An effect evaluation unit is configured to: evaluate the counter-interference effect based on the flight state of the target; Wherein, the state recognition unit comprises: The motion recognition module is configured to: in each unit time slice, use the first sub-classifier to extract the route shortcut change ΔD and the route shortcut standard deviation σ from the motion feature parameters that reflect the horizontal motion state of the target D and the heading angle change Δθ to identify the target's flight action in the horizontal direction, and the second sub-classifier is used to extract the height change Δh and the height standard deviation σ reflecting the vertical motion state of the target from the motion feature parameters h and the standard deviation of height variation σ Δh , to identify the target's flight action in the vertical direction; wherein the unit time slice is obtained by dividing the predetermined time; The state recognition module is configured to: splice the flight actions of multiple unit time slices to output the flight state of the target within the predetermined time.

Images