Predator mechanism based non-lead structure group confrontation method

Abstract

The application discloses a kind of based on predator mechanism's no navigation structure group confrontation method, including steps Step1.Acquire no navigation structure group information;Step2.Analysis no navigation structure group movement model;Step3.Develop predator movement model;Step4.Predator implements the destruction to group structure;Step5.Determine confrontation effect;The above-mentioned method of the application describes the communication characteristics and cooperative behavior characteristics of no navigation structure group by analyzing fish school emergence mechanism and movement model, and a corresponding predator model is designed accordingly, and the cohesion and centripetal property of no navigation structure group are destroyed by the predator, to realize the effective confrontation to the whole no navigation structure group.The predator model proposed in the application is simple, can realize the effective confrontation to no leadership structure group with point-to-surface, and has the advantages of high confrontation efficiency ratio.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of non-leader structure group confrontation, and particularly relates to a non-leader structure group confrontation method based on a predator mechanism. BACKGROUND

[0002] As one of the five intelligent forms of the new generation of artificial intelligence, group intelligence mainly realizes distributed and decentralized intelligent behavior by studying the wisdom of dispersed and self-organizing biological groups, and is usually used for the collaborative work of unmanned aerial vehicle and robot clusters, and has important application prospects in civil and military fields. At the same time, various potential dangers and security risks are increasing day by day, so the research on the confrontation strategy of group intelligence is particularly important.

[0003] The ordered behavior of biological groups emerges deep-seated intelligent connotation, and presents a large number of intelligent behaviors of scale and system through the cooperation between biological individuals. By studying the behavior characteristics of biological group intelligence, the intelligent emergence mechanism contained in the behavior characteristics can provide theoretical and technical support for the research on the confrontation strategy of group intelligence. Based on this, biological groups can be divided into non-leader structure groups and leader-following structure groups according to whether there is a central leader in the group.

[0004] The current confrontation method of non-leader structure group mainly uses technical means to realize the confrontation of the whole group by counteracting a single individual. This confrontation method is difficult to form effective confrontation to group intelligence in actual use. Therefore, in order to efficiently confront the non-leader structure group, theoretically, the key points of counteraction can be analyzed by combining the fish group intelligence emergence mechanism, focusing on the communication characteristics and cooperative behavior characteristics of the group, so as to realize the efficient confrontation of the non-leader structure group. However, from the existing research results, there is no research on the confrontation method of non-leader structure group based on the fish group intelligence emergence mechanism.

[0005] Therefore, it is urgent to design a non-leader structure group confrontation method based on a predator mechanism to solve the problems existing in the prior art. SUMMARY

[0006] In view of the above problems, the present application aims to provide a non-leader structure group confrontation method based on a predator mechanism. The method uses the evasion characteristics of individuals in the group to the predator to step by step disintegrate the communication and cooperative behavior of the non-leader structure group, so as to achieve the purpose of counteraction. Only one predator is needed to effectively counteract the whole non-leader structure group, which can effectively improve the counteraction efficiency ratio, achieve the effect of counteracting from point to surface, and has the characteristics of good counteraction effect and high counteraction efficiency ratio.

[0007] In order to achieve the above object, the technical scheme adopted by the present application is as follows:

[0008] A non-lead structure group confrontation method based on a predator mechanism, comprising the steps of

[0009] Step 1. Obtain non-lead structure group information:

[0010] Video image acquisition is performed on the non-lead structure group, feature extraction is performed on the collected images and videos, and a density distribution map is constructed, thereby obtaining non-lead structure group information;

[0011] Step 2. Determine the non-lead structure group motion model based on the non-lead structure group information:

[0012] Based on the non-lead structure group information, the fish behavior law and emergence mechanism are combined to calculate the corresponding motion vector of each individual in the non-lead structure group, and a non-lead structure group motion model and a structure model are established;

[0013] Step 3. Formulate the predator motion model:

[0014] According to the obtained non-lead structure group motion model, a predator motion model is designed;

[0015] Step 4. Use the predator motion model to implement the destruction of the group structure:

[0016] The predator motion model is used to attack the non-lead structure group, so that the predator successively destroys the group cohesion and centripetal property, and promotes the group to produce dispersed individuals;

[0017] Step 5. Determine the confrontation effect: through real-time monitoring of the predator hunting situation, if the predator is successful in hunting, the group confrontation effect is achieved, if the predator has not eaten for a long time, Step 4 is turned to, and the group center is repositioned.

[0018] Preferably, the process of obtaining non-lead structure group information in step Step 1 comprises:

[0019] Step 101. Video image acquisition of the target group is performed by using visible light sensors, infrared sensors and radars in cooperation;

[0020] Step 102. Feature extraction is performed on the collected images and videos by using a convolutional neural network, and a density distribution map is constructed;

[0021] The convolutional neural network is a hybrid dilated convolutional neural network model.

[0022] Preferably, in the hybrid dilated convolutional neural network model of step Step 102:

[0023] (1) The front end of the hybrid dilated convolutional neural network model consists of the first 10 layers of VGG-16, and the output feature map size is 1 / 8 of the input image. The back end network model consists of 4 cascaded hybrid dilated convolutions, each group consisting of 3 consecutive dilated convolutions. Based on the combination principle of dilation rate, the dilation rate within the group is [1,2,3];

[0024] (2) Taking a drone swarm as an example, in the hybrid dilated convolutional neural network model, the Gaussian kernel density distribution function F(x) and the Gaussian kernel function G in the image of the drone swarm are... σ (x) is constructed by convolution as follows:

[0025]

[0026] In equation (2), x i The pixel values ​​representing the locations of labeled drones, where N represents the number of labeled drones, and δ(xx) i ) represents the impulse function of the drone's pixel position. Let x be the distance i The average distance between the nearest m drones and this drone.

[0027] Preferably, the process of constructing the density distribution map in step 102 includes:

[0028] (1) In the dilated convolutional neural network model, the first 10 layers of VGG-16 are used to complete the feature extraction of the image.

[0029] (2) Use the hybrid dilated convolution strategy to convolve the extracted UAV target pixels in the image to complete the extraction of image features;

[0030] (3) After obtaining the image features, let the location of a labeled drone be at pixel x. i An image contains N labeled drones. The Gaussian kernel density distribution function F(x) of the drone swarm in this image can be obtained by combining it with the Gaussian kernel function G. σ (x) is constructed by convolution as follows:

[0031]

[0032] In equation (2), δ(xx) i ) represents the impulse function of the drone's pixel position. Let x be the distance i The average distance between the nearest m drones and this drone;

[0033] (4) Filter and normalize the image after the target is labeled, and the sum of the filtering results of each labeled point is the density distribution map.

[0034] Preferably, the construction process of the non-structure group motion model in Step 2 is based on the three behavioral characteristics of fish group, i.e. repulsion, parallelism and attraction, and the specific construction process includes

[0035] (1) Repulsion: when the individuals in the group are too close, the individuals will escape in the opposite direction, and the motion vector is:

[0036]

[0037] wherein in formula (3), m represents the number of neighboring individuals, and respectively represent the position of the fish group individual and the i-th neighboring individual of the fish group individual; when the fish group individual approaches the average center position of its neighbors, the individual will show a motion state of moving away from the average center, and the motion direction is

[0038] (2) Parallelism: the individual adjusts its direction to maintain consistency with the motion direction of the neighbor individual, and the motion vector is:

[0039]

[0040] wherein in formula (4), i represents the i-th neighbor individual, represents the current motion direction of the i-th neighbor individual;

[0041] (3) Centripetal property: the individual will try to approach the center of the neighbors according to the position of the neighbor individual to ensure that it will not deviate from the group, and the motion vector points to the average center position of the neighbors of the individual, which is set as

[0042] (4) Cohesion: only centripetal property may produce multiple discontinuous and dispersed small groups, and in order to ensure the consistency of the entire group, the individual needs to approach the center of the entire group, and the motion vector is:

[0043]

[0044] wherein in formula (5), n represents the number of individuals in the fish group;

[0045] (5) Arrangement: in order to ensure that the individuals move in the same direction after forming a group, the individual needs to add a motion vector in the same direction as the group motion direction:

[0046]

[0047] In summary, the motion vector of the individual i in the next moment in the group is:

[0048]

[0049] wherein λ i (i=i, 2, …, 5) is the weight coefficient corresponding to each behavior rule, and the sum of each weight is 1.

[0050] Preferably, the establishment of the predator movement model in Step 3 comprises

[0051] Step 301. Set the predator movement vector:

[0052] To ensure that the predator can successfully destroy the leaderless structure group, it is necessary to ensure that the speed of the predator is not lower than the movement speed of the individual in the leaderless structure group, and the direction of the speed vector changes with the different countermeasures stages;

[0053] Step 302. Set the evasion vector of the individual in the leaderless structure group to the predator as The direction is from the predator to the individual in the group, and different values are taken according to the different hunting stages.

[0054] Preferably, the process of destroying the group structure by using the predator movement model in Step 4 comprises

[0055] Step 401. Destroy the cohesion of the group

[0056] The predator destroys the cohesion of the group by launching an attack on the center of the group, and the has

[0057]

[0058] Step 402. Destroy the centripetal property of the group

[0059] At this time, the predator continues to launch an attack on the center of these small groups, so that the individuals move alone, and the

[0060]

[0061] Step 403. The predator finds the individual who has left the group and hunts it.

[0062] The beneficial effects of the present application are: the present application discloses a leaderless structure group confrontation method based on a predator mechanism, compared with the prior art, the improvement of the present application lies in that

[0063] ​This invention proposes a predator-based method for adversarial behavior in leaderless groups. The method includes the following steps: Step 1. Obtaining information about the leaderless group; Step 2. Analyzing the group's motion model; Step 3. Developing a predator motion model; Step 4. Predators disrupting the group's structure; Step 5. Determining the effectiveness of the adversarial behavior. In practice, this method analyzes the emergence mechanism of fish schools and describes the communication and cooperative behavior characteristics of leaderless groups through motion models, and designs a corresponding predator model. Predators disrupt the cohesion of the leaderless group. This method utilizes predator-prey mechanisms to systematically disrupt the communication and collaborative behavior of unleaderless groups, thereby achieving countermeasures. Furthermore, by employing a predator-prey mechanism to exploit the predator avoidance behavior of individuals within the group, it effectively counteracts the entire unleaderless group, significantly improving the cost-effectiveness of countermeasures and achieving a point-to-area countermeasure effect. Experimental results demonstrate that the proposed predator model is simple, capable of effectively countering unleaderless groups by breaking through their defenses, and possesses the advantages of good countermeasure effectiveness and high cost-effectiveness. Attached Figure Description

[0064] Figure 1 The flowchart of the leaderless group confrontation strategy that introduces the predator mechanism to this invention is shown.

[0065] Figure 2 This is a structural diagram of the hybrid dilated convolutional neural network model of the present invention.

[0066] Figure 3 This is a schematic diagram illustrating the fish swarm countermeasure that disrupts cohesion according to the present invention.

[0067] Figure 4 This is a schematic diagram illustrating the method of countering fish schools that disrupt centripetal orientation, as described in this invention.

[0068] Figure 5 This is a simulation experiment diagram of predators disrupting the cohesion of a fish school according to the present invention.

[0069] Figure 6 This is a simulation diagram of a predator disrupting the centripetal force of a fish school according to the present invention.

[0070] Among them: Figure 5 In the figure, Figure (a) shows the simulation experiment of predator disruption of fish cohesion at time t=42, Figure (b) shows the simulation experiment of predator disruption of fish cohesion at time t=47, Figure (c) shows the simulation experiment of predator disruption of fish cohesion at time t=51, and Figure (d) shows the simulation experiment of predator disruption of fish cohesion at time t=135.

[0071] exist Figure 6In the figures, Fig. (a) is a simulation diagram of the predator destroying the centripetal fish school at time t = 42, Fig. (b) is a simulation diagram of the predator destroying the centripetal fish school at time t = 56, Fig. (c) is a simulation diagram of the predator destroying the centripetal fish school at time t = 69, and Fig. (d) is a simulation diagram of the predator destroying the centripetal fish school at time t = 135. DETAILED DESCRIPTION

[0072] In order for those skilled in the art to better understand the technical solutions of the present application, the technical solutions of the present application will be further described below in combination with the drawings and examples.

[0073] Example 1: Referring to the drawings Figures 1-6 The predator mechanism-based non-lead structure group confrontation method shown in the drawings comprises the steps of

[0074] Step 1. Obtain non-lead structure group information:

[0075] Video images of the non-lead structure group are collected, and the collected images and videos are subjected to feature extraction and construction of a density distribution map, so as to obtain non-lead structure group information;

[0076] Step 101. Video images of the target group are collected by using a multi-sensor system comprising a visible light sensor, an infrared sensor and a radar, etc. In the actual collection process, both the visible light sensor and the infrared sensor are two-dimensional imaging sensors, and the distance information of the target is always missing when the target is positioned in three dimensions, which is an inherent disadvantage of the two types of sensors. In order to supplement the distance information of the target, a radar sensor is used to measure the distance of the target. The three-dimensional positioning information of the target can be obtained through the sensors, and the multi-sensor system has the characteristics of all-weather and all-day, so that the video images and position coordinates of the target group are obtained.

[0077] Step 102. In order to accurately detect the spatial density distribution of the non-lead structure group of the target group video images and position coordinates collected in Step 101:

[0078] (1) A hybrid dilated convolution neural network model is used in this embodiment, and the structure is as shown in Figure 2 The front end of the network model uses the first 10 layers of VGG-16, and the output feature map size is 1 / 8 of the input image. The rear network model is composed of 4 groups of hybrid dilated convolution, each group is composed of 3 consecutive dilated convolutions. Based on the combination principle of dilated rate, the dilated rate in the group is [1, 2, 3]; the VGG network has strong feature extraction capability and is used to extract image features in the density estimation process;

[0079] (2) Because ordinary convolution increases the receptive field and downsamples the image, i.e., reduces the image size, reducing the image size inevitably results in the loss of some information; when dilated convolution is used for calculation, dilated convolution can increase the image receptive field without reducing the image size. The output of dilated convolution is:

[0080]

[0081] In equation (1), x(m,n) represents the input image information with dimensions m and n, which is processed by the convolution kernel w(i,j) to obtain the output of dilated convolution. r is the dilation rate. When r = 1, dilated convolution is a normal convolution. Although dilated convolution can increase the receptive field of the convolution kernel without changing the image size or increasing the network parameters, it also has disadvantages. The larger the dilation rate, the lower the pixel utilization, which reduces the continuity in the image space and results in low correlation between adjacent pixels in the output feature map. This makes it difficult for low-resolution images to function properly. The target drone is more difficult to identify; combined with the segmented dilated convolution mechanism, this embodiment proposes a hybrid dilated convolution strategy. Unlike the constant dilation rate, the hybrid dilated convolution sets several consecutive convolutional layers as a group, and uses convolution with different dilation rates within the group, so that the receptive field of the image has no holes or missing parts, and all pixels participate in the convolution operation, avoiding the loss of some pixel information. The convolution result has continuity in the image space, solving the "grid effect" that will occur in dilated convolution, and effectively improving the utilization rate of drone target pixels in the image during the acquisition process.

[0082] (3) After obtaining the image features, assume that the location of a labeled drone is at pixel x i If there are N labeled drones in an image, then the Gaussian kernel density distribution function F(x) of the drone swarm in the image can be obtained by combining it with the Gaussian kernel function G. σ (x) is constructed by convolution as follows:

[0083]

[0084] In equation (2), δ(xx) i ) represents the impulse function of the drone's pixel position. Let x be the distance i The average distance between the nearest m drones and this drone;

[0085] (4) The image after target labeling is filtered and normalized. The sum of the filtering results of each labeling point is the density distribution map, thus completing the collection and acquisition of information on the group without a navigation structure.

[0086] Step 2: Determine the motion model of the unguided group:

[0087] On the basis of the non-leader structure group information, combining fish behavior rules and emergence mechanism, the corresponding motion vector of each individual in the non-leader structure group is calculated, so as to obtain the whole motion model and structure model of the non-leader structure group;

[0088] The fish behavior rules and emergence mechanism illustrate that the basic motion rules of the non-leader structure group taking fish as an example include repulsion-parallelity-attraction, under the constraint of the rules, the group motion neither produces collision nor loses contact, and there is no "leader" to organize the gathering behavior in the group, but the whole group behavior characteristics of fish wave motion, foraging behavior and migration behavior are emerged through the interaction and coordination between all individuals, so as to construct the motion model of the non-leader structure group;

[0089] The non-leader group shows three behavior characteristics of repulsion, parallelity and attraction in the motion process, that is, the group maintains strong aggregation while avoiding collision, and maintains high consistency in the motion direction and speed, and these properties can be modeled as follows:

[0090] (1) Repulsion: when the individuals in the group are too close, the individuals will escape in the opposite direction, that is, the motion vector is:

[0091]

[0092] In formula (3), m represents the number of neighboring individuals, and respectively represent the position of the fish individual and the i-th neighboring individual of the fish; when the fish individual approaches the average center position of the neighboring individuals, the individual will show the motion state of moving away from the average center, that is, the motion direction is

[0093] (2) Parallelity: the individual adjusts its direction to maintain consistency with the motion direction of the neighboring individuals, that is, the motion vector is:

[0094]

[0095] In formula (4), the i-th neighboring individual represents the current motion direction of the i-th neighboring individual;

[0096] (3) Centripetal property: the individual will try to approach the center of the neighbors according to the position of the neighbors to ensure that it will not deviate from the group, that is, the motion vector points to the average center position of the neighboring individuals, which is set as

[0097] (4) Cohesion: only the centripetal property may produce multiple discontinuous and dispersed small groups, in order to ensure the consistency of the whole group, the individual needs to approach the center of the whole group, that is, the motion vector is:​

[0098]

[0099] wherein in formula (5), n represents the number of individuals in the fish school;

[0100] (5) Arrangement: in order to ensure that the individuals move in the same direction after forming a group, an additional movement vector in the same direction as the group movement direction is added to the individuals:

[0101]

[0102] Therefore, the movement vector of individual i in the next moment in the group is as follows:

[0103]

[0104] wherein λ i (i = 1, 2, …, 5) is the weight coefficient corresponding to each behavior rule, and the sum of the weights is 1; by analyzing the above properties, the movement characteristics of the leaderless structure group can be analyzed, and based on this, a targeted predator movement model can be used to destroy the group movement structure;

[0105] Step 3. Establish a predator movement model:

[0106] According to the obtained leaderless structure group movement model, a predator movement model is designed, including the movement vector of the predator and the predation mechanism; wherein the establishment of the predator movement model comprises:

[0107] Step 301. Set the predator movement vector: in order to ensure that the predator can successfully destroy the leaderless structure group, it is necessary to ensure that the speed of the predator is not lower than the movement speed of the individuals in the leaderless structure group, and the direction of the speed vector changes with the different countermeasures stages;

[0108] Step 302. Set the evasive vector of the individuals in the leaderless structure group to the predator The direction is from the predator to the individual in the group, and according to the different predation stages, different values are taken, and the evasive characteristics of the individuals in the leaderless structure group to the predator can be realized by applying interference means such as sound waves and strong light;

[0109] Step 4. The predator implements the destruction of the group structure: using the predator movement model developed, the predator attacks the leaderless structure group, and uses the density distribution map obtained in Step 1 to locate the predator attack position, so that the predator destroys the cohesion and centripetal property of the group in turn, and promotes the group to produce discrete individuals, facilitating the predation of the predator; wherein the design process of the predator implementing the destruction of the group structure comprises

[0110] Step 401. Destroy the cohesion of the group

[0111] In order to keep the group structure stable and robust, the individuals in the leaderless structure group will spontaneously aggregate to the center, that is, cohesion. According to the density distribution function obtained in Step 102, we can get the point with the maximum density, which is the center of the group. The predator can destroy the cohesion of the group by launching an attack on the center of the group at this time There is

[0112]

[0113] Step 402. Destroy the cohesion of the group

[0114] The individuals in the group not only have cohesion, but also have centripetal property to keep the group in contact with the neighbors. Therefore, when the cohesion of the group is destroyed, the group will be divided into several small groups. Then, the center position of the small group is obtained by using the density distribution diagram at this time. The predator can continue to launch an attack on the center of these small groups to make the individuals move apart at this time

[0115]

[0116] Step 403. The predator can break them one by one by looking for dispersed individuals;

[0117] Step 5. Determine the counteracting effect

[0118] The hunting of the predator is monitored in real time by the monitoring system. If the predator is successful in hunting, the group counteracting effect is achieved. If the predator has not eaten for a long time, Step 4 is turned to try to reposition the center of the group.

[0119] Example 2: Different from example 1, in order to verify the effectiveness of the leaderless group counteracting method based on the predator mechanism as described in example 1, this example takes fish group as an example. First, the motion vectors and weight coefficients of each property are set according to the motion characteristics of the fish group, and the speed of the predator is set to be the same as that of the fish group. The destruction of the cohesion and centripetal property of the leaderless group by the predator is simulated respectively:

[0120] (1) Simulation of destroying the cohesion of the fish group:

[0121] In order to verify the rationality of destroying the cohesion of the leaderless group, the weight coefficients of repulsion, alignment, centripetal property, cohesion and alignment are set to λ1=0.3, λ2=0.1, λ3=0.1, λ4=0.1 and λ5=0.4 respectively according to the fish group motion model and behavior mechanism. As Figure 5As shown in the figure, the X symbol represents the predator, the hexagonal point represents the "food" that has an attractive force to the group individuals, which is set to be fixed at the (50, 50) coordinate position, and the other round points represent the group individuals without the leader structure. In Figure 5 In (a), at t=42, the fish swarm normally forages around the food center. In Figure 5 In (b), at t=47, the X symbol representing the predator appears around the fish swarm, but is not detected by the fish swarm, so the fish swarm still normally forages; in Figure 5 In (c), at t=51, the predator approaches the center of the fish swarm, and the fish swarm individuals close to the predator have already detected the danger and begin to avoid; in Figure 5 In (d), at t=135, the predator completes the attack on the center of the fish swarm, and it can be seen that the fish swarm is no longer a complete group, but is dispersed into several small groups;

[0122] From the simulation results above, it can be seen that the fish swarm, through the mechanisms of target attraction, collision avoidance repulsion, and swarm cooperation, approaches the food while maintaining a stable formation. When the predator appears, the fish swarm, while approaching the food, additionally receives a repulsive force generated by the predator, causing the fish swarm formation to become chaotic and to avoid the predator as much as possible. It can be found that the predator can have a destructive effect on the group, causing the leaderless structure fish swarm to avoid the predator while the group formation shows a chaotic trend. The presence of the predator plays an important role in destroying the cohesion of the leaderless structure and counteracting the overall leaderless structure group;

[0123] (2) Simulation of destroying the centripetal property of the fish swarm

[0124] In order to verify the rationality of destroying the centripetal property of the fish swarm and the fish swarm individuals, the weight coefficients of repulsion, alignment, centripetal property, cohesion, and arrangement are set as λ1=0.3, λ2=0.1, λ3=0.1, λ4=0.4, and λ5=0.1, respectively. As shown in the figure, the X symbol represents the predator, the hexagonal point represents the "food" that has an attractive force to the group individuals, which is set to be fixed at the (50, 50) coordinate position, and the other round points represent the group individuals without the leader structure. In Figure 6 (a), at t=42, the predator is around the food of the fish swarm, and the fish swarm individuals are in an avoidance state, scattered and divided into several nearby groups; in Figure 6 (b), at t=56, the predator starts from the center of the fish swarm and attacks the smaller number of fish swarms, destroying the centripetal property of the fish swarm; in Figure 6 (c), at t=69, the nearby group attacked by the predator is destroyed, and the lone fish swarm individual appears, and the predator attacks the lone fish swarm alone; in Figure 6 (d), at t=135, the predator completes the predation on the lone individual; Figure 6 ​

[0125] The above simulation experiment focuses on the centripetal behavior of fish schools. During the movement of the fish school, the fish school will always move towards the center of the whole. Due to the repulsion effect of predators, the fish school will split into multiple subgroups. While avoiding predators, the subgroups will move towards the average center of the fish school. During this process, multiple independent individuals will appear and will not be able to merge into the whole fish school in time. By preying on the isolated individuals, the fish school will be gradually destroyed and eventually the whole fish school will be successfully countered.

[0126] The data from this embodiment clearly shows that the predator-based unguided group confrontation method described in Embodiment 1 systematically disrupts the communication and collaborative behavior of the unguided group, thereby achieving the purpose of countermeasures. At the same time, it can effectively counter the entire unguided group, effectively improve the cost-effectiveness of countermeasures, and achieve a point-to-area countermeasure effect. It has the characteristics of good countermeasure effect and high cost-effectiveness.

[0127] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A leaderless structure swarm confrontation method based on a predator mechanism, characterized in that: Comprising the steps of Step 1. Obtain the information of the non-leader structure group: Video image collection is performed on the non-leader structure group, feature extraction is performed on the collected images and videos, and a density distribution map is constructed, so as to obtain the information of the non-leader structure group; Step 2. Determine the non-leader structure group motion model according to the non-leader structure group information: On the basis of the non-leader structure group information, the fish school behavior rule and the emergence mechanism are combined to calculate the corresponding motion vector of each individual in the non-leader structure group, and a non-leader structure group motion model and a structure model are established; Step 3. Formulate the motion model of the predator: According to the obtained non-leader structure group motion model, the motion model of the predator is designed; Step 4. Use the predator motion model to implement the destruction of the group structure: The predator motion model is used to attack the non-leader structure group, so that the predator successively destroys the cohesion and centripetal property of the group, and promotes the group to produce discrete individuals; Step 5. Determine the counteracting effect: through real-time monitoring of the hunting situation of the predator, if the predator is successful in hunting, the group counteracting effect is achieved, if the predator has not eaten for a long time, Step 4 is turned to, and the group center is repositioned.

2. The method of claim 1, wherein the method is a leaderless swarm method based on a predator mechanism. The process of obtaining the information of the non-leader structure group in Step 1 comprises: Step 101. Video image collection is performed on the target group by using visible light sensors, infrared sensors and radars in cooperation; Step 102. Feature extraction is performed on the collected images and videos by using a convolutional neural network, and a density distribution map is constructed; The convolutional neural network is a hybrid dilated convolutional neural network model.

3. The leaderless swarm confrontation method based on predator mechanism according to claim 2, characterized in that: In the hybrid dilated convolutional neural network model of Step 102: (1) The front end of the hybrid dilated convolutional neural network model is the first 10 layers of VGG-16, and the output feature map size is 1 / 8 of the input image, and the rear network model is composed of 4 groups of hybrid dilated convolution in series, each group is composed of 3 consecutive dilated convolutions, and the group internal dilated rate is [1, 2, 3] based on the combination principle of dilated rate; (2) Taking the UAV group as an example, in the mixed cavity convolution neural network model, the Gaussian kernel density distribution function of the UAV group in the image is convoluted to construct: and the Gaussian kernel function ​ （2） wherein in formula (2), represents the pixel points of the labeled UAV position, N represents the number of labeled UAVs, represents the pulse function of the UAV pixel position, , is the distance average distance of the nearest m UAVs.

4. The leaderless swarm confrontation method based on predator mechanism according to claim 3, characterized in that: The process of constructing the density distribution map in Step 102 comprises (1) In the dilated convolutional neural network model, the first 10 layers of VGG-16 are used to complete feature extraction of the image; (2) The hybrid dilated convolution strategy is used to convolve the UAV target pixel points in the extracted image, and the feature extraction of the image is completed; (3) After obtaining the image features, assume that the location of a labeled UAV is at pixel point The image contains N labeled drones. The Gaussian kernel density distribution function of the drone swarm in the image is given. It can be compared with the Gaussian kernel function The convolution structure is as follows: （2） wherein in formula (2), a pulse function representing the drone pixel position, , is the distance average distance of the m closest drones; (4) The target labeled image is filtered and normalized to obtain the sum of the filtering results of each labeled point, which is the density distribution map.

5. The leaderless swarm confrontation method based on predator mechanism according to claim 1, characterized in that: Therefore, the construction process of the non-leader structure group motion model in Step 2 is based on the repulsion, parallelism and attraction of the fish school, and the specific construction process comprises (1) Repulsion: when the individuals in the group are too close, the individuals will escape in the opposite direction, and the motion vector is: （3） wherein in formula (3), m represents the number of neighboring individuals, and respectively represent the position of a fish individual and its mth neighboring individual; when a fish individual approaches the average center position of its neighbors, the individual will exhibit a movement state away from the average center, in the direction ; and ; (2) Parallelism: the individual adjusts its direction to keep consistent with the motion direction of the neighbor individual, and the motion vector is: （4） wherein in formula (4), represents the current movement direction of the neighbor individual. (3) Centripetal: an individual will try to move towards the center of its neighbors to ensure that it does not leave the group, the motion vector is directed towards the average center position of the individual's neighbors, set as ; (4) Cohesion: only centripetal property may produce multiple discontinuous and dispersed small groups, in order to ensure the consistency of the whole group, the individual needs to move towards the center of the whole group, and the motion vector is: （5） Wherein in formula (5), n represents the number of individuals in the fish school; (5) Arrangement: to ensure that the individual forms a group, the individual needs to add a movement vector in the same direction as the group movement direction: （6） In summary, the next moment movement vector of individual i in the group is: （7） wherein are weight coefficients corresponding to each behavior rule, and the sum of each weight is 1.

6. The method of claim 1, wherein: Therefore, the establishment of the predator movement model described in step Step3 includes Step301. Set the predator movement vector: To ensure that the predator can successfully destroy the non-leader structure group, it is necessary to ensure that the speed of the predator is not lower than the movement speed of the individuals in the non-leader structure group, and the speed vector direction changes with the different stages of countermeasures; Step 302. Let the evasion vector of an individual in the leaderless structure population towards the predator be which points in the direction from the individual to the predator, and takes different values depending on the stage of the predation. depending on the stage of the predation.

7. The method of claim 1, wherein: Therefore, the process of using the predator movement model to implement the destruction of the group structure described in step Step4 includes Step401. Destroy the cohesion of the group Predators disrupt the cohesion of the group by launching a charge towards the center of the group, against have （8）； Step402. Destroy the centripetal nature of the group At this time, the predator continues to launch an attack on the center of these small groups, making individuals move alone, and （9）； Step403. The predator finds individuals who have left the group and hunts them.

Images