Unmanned aerial vehicle emergency defense control system and method based on deep learning

Abstract

The invention relates to the field of unmanned aerial vehicle control, in particular to an unmanned aerial vehicle emergency defense control system and method based on deep learning, potential intrusion is predicted by using abnormal state data around a deep neural network learning base area, an unmanned aerial vehicle cluster inspection strategy is set, and unmanned aerial vehicle monitoring is quickly initiated in the initial stage of intrusion; local interference of the unmanned aerial vehicle cluster in the inspection process is identified, the flight layout of the unmanned aerial vehicle cluster and the control strategy for at least part of unmanned aerial vehicles are adjusted, and it is ensured that the unmanned aerial vehicles with good performance maintain normal flight of the whole cluster; according to the relative space relation dynamic data between the invader and the unmanned aerial vehicle cluster, the flight state of the unmanned aerial vehicle is adjusted, and accurate enclosing prevention and control of the invader by the unmanned aerial vehicle cluster are achieved; and according to the flight path of the invader, the unmanned aerial vehicle cluster is controlled to initiate non-contact interference on the invader until the flight state of the invader changes, so that flight change scenes of different invader can be effectively adapted, and the defense reliability and timeliness are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of UAV control, and particularly to a UAV emergency defense control system and method based on deep learning. Background Art

[0002] UAVs have the characteristics of miniaturization, convenient operation, flexible flight, and large inspection range, and have become the main tools for performing air surveillance and defense tasks. Existing methods for using UAVs to perform defense tasks mainly focus on monitoring the flight path of invading flying objects and then controlling the UAVs to track the invading flying objects. The above methods have great passivity and uncertainty. Once the flight path of the invading flying object has large variations and the invading flying object actively launches interference attacks on the UAVs, it will cause the UAVs to be unable to change their flight trajectories in a timely and accurate manner, and cannot achieve close-range tracking and containment control of the invading flying objects, seriously affecting the prevention and control efficiency and accuracy of the invading flying objects. It can be seen that how to adaptively adjust the flight layout of the UAV cluster and actively launch interference against the invading flying objects during the whole process of containing and preventing the invading flying objects in case of an invasion event is of great significance for improving the accuracy and timeliness of invasion recognition and quickly controlling the invading flying objects. Summary of the Invention

[0003] Aiming at the defects existing in the prior art, the present invention provides a UAV emergency defense control system and method based on deep learning. By using a deep neural network to learn the abnormal state data around the base area to predict potential invasions and set the inspection strategy for the UAV cluster around the base area, UAV surveillance is quickly launched at the initial stage of the invasion; the interference received locally by the UAV cluster during the inspection process is monitored and identified, and the flight layout of the UAV cluster and the control strategy for at least some of the UAVs are adjusted to ensure that the UAVs with good performance maintain the normal flight of the whole cluster; according to the dynamic data of the relative spatial relationship between the invading object and the UAV cluster, the flight state of the UAVs is adjusted to achieve accurate containment and prevention of the invading object by the UAV cluster, effectively restricting the flight activities of the invading object; and according to the flight trajectory of the invading object, the UAV cluster is controlled to launch non-contact interference against the invading object until the flight state of the invading object changes, which can effectively adapt to different flight change scenarios of the invading object, accurately expel the invading object actively, and improve the reliability and timeliness of defense.

[0004] The present invention provides a UAV emergency defense control method based on deep learning, including: Step S1: Perform deep neural network learning on the abnormal state data around the base area to predict potential invasion events, so as to set the inspection strategy for the UAV cluster around the base area; Step S2: Monitor and identify the interference received locally by the UAV cluster during the inspection process, and adjust the flight layout of the UAV cluster and the control strategy for at least some of the UAVs; Step S3, adjust the flight states of the corresponding drones in the drone cluster according to the dynamic data of the relative spatial relationship between the intruders around the base area and the drone cluster, so that the intruders and the drone cluster meet the preset position relationship conditions; Step S4, according to the flight trajectory of the intruder, control at least one drone in the drone cluster to initiate non-contact interference on the intruder until the flight state of the intruder changes.

[0005] In an embodiment disclosed in the present application, before performing deep neural network learning on the abnormal state data around the base area in the step S1, it includes: Perform visual scanning detection and radar scanning detection on the periphery of the base area to obtain the peripheral environment image and peripheral environment radar data of the base area; Analyze the peripheral environment radar data to determine the flight dynamic path of the suspected foreign objects around the base area; Based on the flight dynamic path, perform local recognition on the peripheral environment image to obtain the flight action postures of the suspected foreign objects; Perform time-domain evolution analysis on the flight dynamic path and flight action postures of the suspected foreign objects to obtain the abnormal state data around the base area; wherein, the abnormal state data includes the flight dynamic path data and flight action posture data of the suspected foreign objects approaching the base area.

[0006] In an embodiment disclosed in the present application, in the step S1, perform deep neural network learning on the abnormal state data around the base area to predict potential intrusion events, including: After performing noise removal and reliable data screening processing on the flight dynamic path data and flight action posture data included in the abnormal state data, input the flight dynamic path data and the flight action posture data into a deep neural network model for analysis and learning to obtain the flight trend of the suspected foreign objects relative to the base area; According to the flight trend, predict the spatial distribution of potential intrusion events in the base area.

[0007] In an embodiment disclosed in the present application, in the step S1, set the inspection strategy of the drone cluster around the base area, including: According to the spatial distribution of potential intrusion events in the base area and the geographical terrain distribution around the base area, identify the intruders around the base area and the spatial distribution of their intrusion locations; Set the inspection strategy for the UAV cluster around the base area according to the spatial distribution of the intrusion locations and the readiness status of the UAV clusters within the base area; wherein, the readiness status of the UAV clusters includes the parking positions and takeoff acceleration characteristics of all the UAV clusters within the base area; and the inspection strategy for the UAV clusters includes the inspection flight coverage spatial range of the UAV clusters around the base area.

[0008] In an embodiment disclosed in the present application, in the step S2, monitoring and identifying the interference suffered by the UAV cluster locally during the inspection process includes: Listen to the external signals received locally by each UAV under the UAV cluster during the inspection process, and extract the interference signal components of the external signals; Monitor the flight trajectory and attitude drift change data of each UAV under the UAV cluster during the inspection process, analyze the flight trajectory and attitude drift change data, and obtain the occurrence time of the abnormal UAV flight; Compare the time series of the interference signal components in the external signals with the occurrence time of the abnormal UAV flight to identify the UAVs in the UAV cluster that have experienced flight instability events.

[0009] In an embodiment disclosed in the present application, in the step S2, adjusting the flight layout of the UAV cluster and the control strategy for at least some of the UAVs includes: According to the distribution position of the UAVs that have experienced flight instability events within the UAV cluster and the relative flight speed between the UAVs that have experienced flight instability events and other UAVs, adjust the flight layout of the UAV cluster, switch the UAVs that have experienced flight instability events to the internal area of the UAV cluster, and switch other UAVs to the peripheral area of the UAV cluster; According to the communication control strategy between the UAVs that have experienced flight instability events and the ground base station, switch the UAVs that have experienced flight instability events to the working state controlled by other UAVs, so that other UAVs perform positioning and flight control on the UAVs that have experienced flight instability events.

[0010] In an embodiment disclosed in the present application, in the step S3, it further includes: Obtain the predicted flight trajectory of the intruders around the base area and the planned flight trajectory of the entire UAV cluster; Spatially compare the predicted flight trajectory with the planned flight trajectory to obtain dynamic data on the relative spatial relationship between the intruders around the base area and the UAV cluster; wherein, the dynamic data on the relative spatial relationship includes the relative spatial distance and the dynamic data on the relative spatial orientation between the intruders around the base area and the UAV cluster.

[0011] In an embodiment disclosed in the present application, in the step S3, according to the dynamic data on the relative spatial relationship between the intruders around the base area and the UAV cluster, adjust the flight states of the corresponding UAVs in the UAV cluster, so that the intruders and the UAV cluster meet the preset position relationship conditions, including: Perform a time-evolution analysis on the dynamic data of the relative spatial distance and the relative spatial orientation between the intruders around the base area and the UAV cluster to predict the range of the encounter positions between the UAV cluster and the intruders; According to the range of the encounter positions, adjust the flight trajectories and / or flight speeds of the UAVs in the outer area of the UAV cluster, so that the UAV cluster can form an encirclement of the intruders.

[0012] In an embodiment disclosed in the present application, in the step S4, according to the flight trajectory of the intruder, control at least one UAV in the UAV cluster to initiate non-contact interference on the intruder until the flight state of the intruder changes, including: Determine the UAV in the UAV cluster that is closest to the intruder according to the flight trajectory of the intruder; According to the characteristics of the relative azimuth change between the closest UAV and the intruder, control the operation intensity and operation duration of the closest UAV to initiate electromagnetic interference and / or laser interference on the intruder until the intruder flies away from the base area or stops flying.

[0013] The present invention also provides a UAV emergency defense control system for deep learning, including: A deep learning and inspection strategy setting module, which is used to perform deep neural network learning on the abnormal state data around the base area, predict potential intrusion events, and thus set the inspection strategy for the UAV cluster around the base area; An interference monitoring and identification module, which is used to monitor and identify the interference received by the UAV cluster locally during the inspection process; A flight layout and control strategy adjustment module, which is used to adjust the flight layout of the UAV cluster and the control strategy for at least some of the UAVs; The cluster flight state adjustment module is used to adjust the flight states of the corresponding drones in the drone cluster according to the dynamic data of the relative spatial relationship between the intruders around the base area and the drone cluster, so that the intruders and the drone cluster meet the preset position relationship conditions; The interference initiation control module is used to control at least one drone in the drone cluster to initiate non-contact interference on the intruder according to the flight trajectory of the intruder until the flight state of the intruder changes.

[0014] Compared with the prior art, the deep learning-based drone emergency defense control system and method use a deep neural network to learn the abnormal state data around the base area to predict potential intrusions and set the inspection strategies of the drone cluster, and quickly initiate drone surveillance at the initial stage of the intrusion; identify the interference suffered by the drone cluster locally during the inspection process, adjust the flight layout of the drone cluster and the control strategies for at least some drones to ensure that the drones with good performance maintain the normal flight of the entire cluster; adjust the drone flight states according to the dynamic data of the relative spatial relationship between the intruders and the drone cluster to achieve accurate containment and prevention of the intruders by the drone cluster, effectively restricting the flight activities of the intruders; also control the drone cluster to initiate non-contact interference on the intruders according to the flight trajectories of the intruders until the flight states of the intruders change, which can effectively adapt to different flight change scenarios of the intruders, accurately expel the intruders actively, and improve the reliability and timeliness of defense.

[0015] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be learned by practicing the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the written specification, claims, and drawings.

[0016] The technical solutions of the present invention will be further described in detail below through the drawings and embodiments. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flowchart of the deep learning-based drone emergency defense control method provided by the present invention.

[0019] Figure 2 It is a schematic diagram of the adjustment of the flight layout of the drone cluster.

[0020] Figure 3 Schematic diagram of non-contact interference initiated by a drone swarm against an intruder.

[0021] Figure 4 Schematic diagram of the structure of the deep learning-based drone emergency defense control system provided by the present invention. Detailed implementation manners

[0022] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0023] Refer to Figure 1 , which is a schematic flowchart of the deep learning-based drone emergency defense control method provided by the embodiment of the present invention. The deep learning-based drone emergency defense control method includes: Step S1: Perform deep neural network learning on the abnormal state data around the base area to predict potential intrusion events, so as to set the inspection strategy for the drone swarm around the base area; Step S2: Monitor and identify the interference received locally by the drone swarm during the inspection process, and adjust the flight layout of the drone swarm and the control strategy for at least some of the drones; Step S3: Adjust the flight states of the corresponding drones in the drone swarm according to the dynamic data of the relative spatial relationship between the intruder around the base area and the drone swarm, so that the intruder and the drone swarm meet the preset position relationship conditions; Step S4: Control at least one drone in the drone swarm to initiate non-contact interference against the intruder according to the flight trajectory of the intruder until the flight state of the intruder changes.

[0024] The beneficial effects of the above technical solution are as follows: The UAV emergency defense control method based on deep learning uses a deep neural network to learn the abnormal state data around the base area to predict potential intrusions and set the UAV cluster patrol strategy, and quickly initiates UAV surveillance at the initial stage of the intrusion; it identifies the interference suffered by the UAV cluster locally during the patrol process, adjusts the flight layout of the UAV cluster and the control strategy for at least some of the UAVs to ensure that the UAVs with good performance maintain the overall normal flight of the cluster; according to the dynamic data of the relative spatial relationship between the intruder and the UAV cluster, it adjusts the flight state of the UAVs to achieve the accurate containment and prevention of the intruder by the UAV cluster, effectively restricting the flight activities of the intruder; it also controls the UAV cluster to initiate non-contact interference on the intruder according to the flight trajectory of the intruder until the flight state of the intruder changes, which can effectively adapt to different flight change scenarios of the intruder, accurately actively expel the intruder, and improve the reliability and timeliness of the defense.

[0025] Preferably, in step S1, before performing deep neural network learning on the abnormal state data around the base area, it includes: Performing visual scanning detection and radar scanning detection on the periphery of the base area to obtain the peripheral environment image and peripheral environment radar data of the base area; Analyzing the peripheral environment radar data to determine the flight dynamic path of suspected foreign objects around the base area; Based on the flight dynamic path, performing local recognition on the peripheral environment image to obtain the flight action postures of suspected foreign objects; Performing time-domain evolution analysis on the flight dynamic path and flight action postures of suspected foreign objects to obtain the abnormal state data around the base area; wherein, the abnormal state data includes the flight dynamic path data and flight action posture data of suspected foreign objects approaching the base area.

[0026] Base areas with spatial openness characteristics such as sports fields and parks are vulnerable to external intrusion surveillance. In order to timely detect the possible intrusion surveillance of the base area, cameras and radar sensors can be arranged at different positions within the base area, and the cameras and radar sensors are used to perform panoramic visual scanning detection and radar scanning detection on the surrounding environment of their respective positions to obtain the peripheral environment image and peripheral environment radar data of the base area. Specifically, the surrounding environment sub-images and surrounding environment radar sub-data generated by each camera and each radar sensor can be collected first, and then according to the distribution position coordinates of all cameras in the base area, all the surrounding environment sub-images generated by the cameras are integrated into the surrounding environment image, and according to the distribution position coordinates of all radar sensors in the base area, all the surrounding environment radar sub-data generated by the radar sensors are integrated into the surrounding environment radar data, to characterize the distribution dynamics of foreign objects in the peripheral environment of the base area at the visible vision and radar vision levels.

[0027] Considering that radar sensors have a longer detection range for objects compared to cameras, when there are foreign objects in the surrounding environment of the base area, the radar sensors will detect the presence of foreign objects earlier than the cameras. At the same time, the radar sensors can also monitor the foreign objects in real time and dynamically to obtain the movement of the foreign objects. Therefore, first extract several radar sub-data corresponding to several time points under the radar data of the surrounding environment, perform time evolution analysis on the several radar sub-data, and determine the flight dynamic path of the suspected foreign objects around the base area. The flight dynamic path of the suspected foreign objects characterizes the spatial range where there are suspected foreign objects around the base area. Then, based on the spatial range where the flight dynamic path of the suspected foreign objects is located, extract the local image fragments where the suspected foreign objects exist from the surrounding environment images, and perform object contour recognition on the above local image fragments to obtain the flight action postures of the suspected foreign objects. By combining the radar sensors and cameras, the foreign objects can be quickly located and dynamically tracked and recognized when the distance between the foreign objects and the base area is far, and the flight dynamic path can be obtained. When the distance between the foreign objects and the base area is close, local image recognition is performed to comprehensively and accurately determine the flight action postures of the foreign objects, providing a reliable basis for subsequent judgment on whether the foreign objects constitute an abnormal impact on the surrounding area of the base area.

[0028] In actual monitoring, not all foreign objects around the base area will eventually invade the base area. Some foreign objects may interfere with the surrounding area of the base area during the flight along the pre-planned path, and these foreign objects should not be considered as potential invaders of the base area. Another part of the foreign objects may invade the base area by approaching gradually. In order to avoid considering all foreign objects around the base area as invaders and imposing a huge defense task pressure on the UAV cluster, it is necessary to screen and select the abnormal state data around the base area. Specifically, perform time domain evolution analysis on the flight dynamic path and flight action postures of the suspected foreign objects, identify the suspected foreign objects that are gradually approaching the base area, and use the flight dynamic path data and flight action posture data of the above suspected foreign objects as the abnormal state data around the base area to facilitate the subsequent further determination of potential intrusion events around the base area.

[0029] Preferably, in step S1, perform deep neural network learning on the abnormal state data around the base area to predict potential intrusion events, including: After performing noise removal and reliable data screening on the flight dynamic path data and flight action posture data included in the abnormal state data, input the flight dynamic path data and flight action posture data into the deep neural network model for analysis and learning to obtain the flight trend of the suspected foreign objects relative to the base area; According to the flight trend, predict the spatial distribution of potential intrusion events in the base area.

[0030] The abnormal state data includes the flight dynamic path data and flight action attitude data of suspected foreign objects near the base area. Considering the large uncertainty in the flight of suspected foreign objects, in order to accurately predict the flight trend of suspected foreign objects relative to the base area, after noise removal and reliable data screening processing of the flight dynamic path data and flight action attitude data included in the abnormal state data, the flight dynamic path data and flight action attitude data are input into the deep neural network model for analysis and learning to comprehensively and accurately predict the flight trend of suspected foreign objects relative to the base area; among them, the above flight trend may include but is not limited to the flight distance change trend and flight azimuth change trend of suspected foreign objects relative to the base area. Then, according to the flight trend of suspected foreign objects relative to the base area, it is judged whether the current flight action of suspected foreign objects constitutes a potential intrusion event on the base area, so as to predict the spatial distribution of potential intrusion events occurring in the base area; among them, the above spatial distribution may be but is not limited to the spatial position range where potential intrusion events occur relative to the base area.

[0031] Preferably, in step S1, a drone cluster inspection strategy for the periphery of the base area is set, including: According to the spatial distribution of potential intrusion events in the base area and the geographical terrain distribution around the base area, identify the intruders around the base area and their spatial distribution of intrusion locations. According to the spatial distribution of intrusion locations and the readiness status of drone clusters in the base area, set a drone cluster inspection strategy for the periphery of the base area; among them, the readiness status of drone clusters includes the parking positions and takeoff acceleration characteristics of all drone clusters in the base area; the drone cluster inspection strategy includes the flight coverage space range of drone clusters during inspection around the base area.

[0032] When it is determined that a potential intrusion event occurs in the base area, it is necessary to trigger the takeoff of the drone cluster to intercept the above potential intrusion event. In order to ensure the immediacy of the takeoff and interception of the drone cluster, it is necessary to select a suitable drone cluster from the base area to perform the takeoff and interception task. In addition, the potential intrusion events in the base area are also affected by geographical terrain factors such as hills around the base area, so that the intruders initiating potential intrusion events need to bypass the hill obstacles to enter the base area. In order to accurately determine the actual intruders and intrusion locations around the base area, it is necessary to make a spatial comparison between the spatial distribution of potential intrusion events in the base area and the geographical terrain obstacle distribution such as hills around the base area, and identify the intruders that actually cause interference to the periphery of the base area and their spatial distribution of intrusion locations.

[0033] To ensure that the UAV cluster on standby within the base area can respond to the intrusion of intruders in a timely manner, based on the spatial distribution of intrusion locations around the base area and the respective parking positions and takeoff acceleration characteristics of all UAV clusters within the base area, select and determine the UAV clusters with a relatively short distance between the parking position and the intrusion location and a relatively high takeoff acceleration, and set the inspection flight coverage space range around the ground base area for the selected UAV clusters, ensuring that the UAV clusters can take off in a timely manner and reach the above-mentioned inspection flight coverage space range to effectively track and intercept the intruders.

[0034] Preferably, in step S2, monitor and identify the interference received locally by the UAV cluster during the inspection process, including: Listen to the external signals received locally by each UAV under the UAV cluster during the inspection process, and extract the interference signal components of the external signals; Monitor the flight trajectory and attitude drift change data of each UAV under the UAV cluster during the inspection process, analyze the flight trajectory and attitude drift change data, and obtain the time of occurrence of abnormal UAV flight; Compare the time series of the interference signal components in the external signals with the time of occurrence of abnormal UAV flight to identify the UAVs in the UAV cluster that have experienced flight instability events.

[0035] During the process of an intruder approaching the base area, electromagnetic interference may be launched against the UAV cluster, affecting the normal flight of at least some of the UAVs in the UAV cluster. There are differences in the interference situations of the electromagnetic interference launched by the intruder against different UAVs in the UAV cluster. Some UAVs are subject to greater electromagnetic interference, while others are subject to less electromagnetic interference. For the UAVs subject to greater electromagnetic interference, flight failures and inability to communicate normally with the ground base station may occur. To accurately identify the UAVs that are affected by greater interference and cannot fly stably, first listen to the external signals received locally by each UAV under the UAV cluster during the inspection process, and identify the useful command signal components and interference signal components from the external signals; also listen to and analyze the flight trajectory and attitude drift change data of each UAV under the UAV cluster during the inspection process. If the flight trajectory drift amount or flight attitude drift amount of a UAV exceeds the preset drift amount threshold during a certain time interval of the inspection process, then determine the above time interval as the time of occurrence of abnormal UAV flight. Then compare the time series of the interference signal components in the external signals with the time of occurrence of abnormal UAV flight. If there is an overlap between the above time series and the time of occurrence of abnormal UAV flight, then determine that the UAV has experienced a flight instability event; otherwise, determine that the UAV has not experienced a flight instability event, which is convenient for timely and accurately adjusting the position of the UAVs that have experienced flight instability events within the cluster in the future and minimizing the interference impact on the UAVs.

[0036] Preferably, in step S2, adjusting the flight layout of the UAV cluster and the control strategy for at least some of the UAVs includes: According to the distribution position of the UAVs with flight instability events in the UAV cluster and the relative flight speeds between the UAVs with flight instability events and other UAVs, adjusting the flight layout of the UAV cluster, switching the UAVs with flight instability events to the inner area of the UAV cluster, and switching other UAVs to the outer area of the UAV cluster; According to the communication control strategy between the UAVs with flight instability events and the ground base station, switching the UAVs with flight instability events to the working state controlled by other UAVs, so that other UAVs perform positioning and flight control on the UAVs with flight instability events.

[0037] Under the interference of the intrusion object, the UAVs with flight instability events will be unable to position and communicate with the ground base station, resulting in the UAVs being unable to maintain stable flight along the predetermined trajectory and prone to risks such as falling or colliding with other UAVs in the cluster. To reduce the interference impact of the intrusion object on the above UAVs, the flight layout of the UAV cluster is adjusted. As Figure 2 shown, it is the UAV distribution before and after the flight layout adjustment of the UAV cluster, where UAV1 is the UAV with a flight instability event. Before the flight layout adjustment, all UAVs in the UAV cluster can be arranged in a circular array; after the flight layout adjustment, the UAVs with instability events will be switched to the inner area of the UAV cluster (i.e., the area inside the dotted arc), and other UAVs will be switched to the outer area of the UAV cluster (i.e., the area outside the dotted arc), which can physically isolate the UAVs with instability events. Specifically, according to the distribution position of the UAVs with flight instability events in the UAV cluster and the relative flight speeds between the UAVs with flight instability events and other UAVs, the UAVs with flight instability events are switched to the inner area of the UAV cluster, and other UAVs are switched to the outer area of the UAV cluster to ensure that there will be no collision between UAVs during the flight layout adjustment of the UAV cluster. In addition, in order to enable the UAVs with flight instability events to still obtain positioning signals and flight control command signals after switching positions, according to the communication control strategy between the UAVs with flight instability events and the ground base station, the UAVs with flight instability events are switched to the working state controlled by other UAVs to avoid the communication between the UAVs with flight instability events and the ground base station being interfered by the outside world, and instead rely on other UAVs to perform positioning and flight control on the UAVs with flight instability events.

[0038] Preferably, in step S3, it further includes: Obtain the predicted flight trajectory of the intruder around the base area and the planned flight trajectory of the entire UAV cluster; Perform a spatial comparison between the predicted flight trajectory and the planned flight trajectory to obtain dynamic data on the relative spatial relationship between the intruder around the base area and the UAV cluster; among them, the dynamic data on the relative spatial relationship includes the relative spatial distance and the dynamic data on the relative spatial orientation between the intruder around the base area and the UAV cluster.

[0039] In order to track the intruder around the base area in real time and perform precise containment control, it is necessary to control the UAV cluster to form an effective physical space enclosure around the intruder when the relative spatial relationship between the UAV cluster and the intruder matches during the flight process. Specifically, by comparing the predicted flight trajectory of the intruder around the base area with the planned flight trajectory of the entire UAV cluster, dynamic data on the relative spatial distance and the relative spatial orientation between the intruder around the base area and the UAV cluster are obtained, providing a reliable spatial position reference for subsequent control of the UAVs to contain the intruder.

[0040] Preferably, in step S3, according to the dynamic data on the relative spatial relationship between the intruder around the base area and the UAV cluster, adjust the flight states of the corresponding UAVs in the UAV cluster, so that the intruder and the UAV cluster meet the preset position relationship conditions, including: Perform a time-evolution analysis on the dynamic data of the relative spatial distance and the relative spatial orientation between the intruder around the base area and the UAV cluster to predict the range of the encounter position between the UAV cluster and the intruder; According to the range of the encounter position, adjust the flight trajectories and / or flight speeds of the UAVs in the outer area of the UAV cluster, so that the UAV cluster can form an enclosure around the intruder.

[0041] Perform a time-evolution analysis on the dynamic data of the relative spatial distance and the relative spatial orientation between the intruder around the base area and the UAV cluster to predict the range of the encounter position between the UAV cluster and the intruder, and precisely narrow down the space where the UAV cluster is allowed to contain the intruder. Then, according to the range of the encounter position, adjust the flight trajectories and / or flight speeds of the UAVs in the outer area of the UAV cluster, so that the UAV cluster can form an enclosure around the intruder, enabling the UAV cluster to precisely form a physical enclosure isolation around the intruder during the encounter process, facilitating subsequent interference with the intruder and effectively improving the prevention and control efficiency of the intruder.

[0042] Preferably, in step S4, according to the flight trajectory of the intruder, control at least one UAV in the UAV cluster to initiate non-contact interference with the intruder until the flight state of the intruder changes, including: According to the flight trajectory of the intruder, determine the UAV in the UAV cluster that is closest to the intruder; According to the relative azimuth change characteristics between the closest drone and the intruder, control the operation intensity and operation duration of the closest drone to launch electromagnetic interference and / or laser interference against the intruder until the intruder flies away from the base area or stops flying.

[0043] After the drone swarm forms an encirclement of the intruder, the drone swarm will adjust its own flight trajectory according to the flight trajectory of the intruder, so that the drone swarm can form an effective encirclement of the intruder within a certain time length. At this time, the drones with flight instability events in the internal area of the drone swarm can also ignore the interference effects from the intruder after switching the control strategy. At this time, determine the drone with the closest distance between the intruders in the drone swarm, and combine the relative azimuth change characteristics between the closest drone and the intruder, and control the operation intensity and operation duration of the closest drone to launch electromagnetic interference and / or laser interference against the intruder until the intruder flies away from the base area or stops flying. The process of launching non-contact interference against the intruder is as Figure 3 shown. By the above method, the intruder can be quickly sniped and controlled, and the intrusion impact of the intruder on the base area can be reduced.

[0044] Refer to Figure 4 , which is a schematic structural diagram of the deep learning drone emergency defense control system provided by the embodiment of the present invention. The deep learning drone emergency defense control system includes: A deep learning and inspection strategy setting module, which is used to perform deep neural network learning on the abnormal state data around the base area, predict potential intrusion events, and thus set the inspection strategy for the drone swarm around the base area; An interference monitoring and identification module, which is used to monitor and identify the interference received locally by the drone swarm during the inspection process; A flight layout and control strategy adjustment module, which is used to adjust the flight layout of the drone swarm and the control strategy for at least some drones; A cluster flight state adjustment module, which is used to adjust the flight states of the corresponding drones in the drone swarm according to the dynamic data of the relative spatial relationship between the intruders around the base area and the drone swarm, so that the intruders and the drone swarm meet the preset position relationship conditions; An interference launch control module, which is used to control at least one drone in the drone swarm to launch non-contact interference against the intruder according to the flight trajectory of the intruder until the flight state of the intruder changes.

[0045] The operation and effect of the deep learning drone emergency defense control system of the present invention correspond to those of the above-mentioned deep learning drone emergency defense control method, and the deep learning drone emergency defense control system will not be repeated here.

[0046] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims

1. A method for emergency defense control of a drone using deep learning, characterized in that, Including: Step S1: Conduct deep neural network learning on the abnormal state data around the base area to predict potential intrusion events, thereby setting the inspection strategy for the drone cluster around the base area; Step S2: Monitor and identify the interference received locally by the drone cluster during the inspection process, and adjust the flight layout of the drone cluster and the control strategy for at least some of the drones; Step S3: Adjust the flight states of the corresponding drones in the drone cluster according to the dynamic data of the relative spatial relationship between the intruders around the base area and the drone cluster, so that the intruders and the drone cluster meet the preset position relationship conditions; Step S4: According to the flight trajectory of the intruder, control at least one drone in the drone cluster to initiate non-contact interference on the intruder until the flight state of the intruder changes.

2. The deep learning-based drone emergency defense control method according to claim 1, wherein: Before conducting deep neural network learning on the abnormal state data around the base area in the step S1, it includes: Conduct visual scanning detection and radar scanning detection on the periphery of the base area to obtain the peripheral environment image and peripheral environment radar data of the base area; Analyze the peripheral environment radar data to determine the flight dynamic path of the suspected foreign objects around the base area; Based on the flight dynamic path, conduct local recognition on the peripheral environment image to obtain the flight action postures of the suspected foreign objects; Conduct time-domain evolution analysis on the flight dynamic path and flight action postures of the suspected foreign objects to obtain the abnormal state data around the base area; wherein, the abnormal state data includes the flight dynamic path data and flight action posture data of the suspected foreign objects approaching the base area.

3. The deep learning-based drone emergency defense control method according to claim 2, wherein: When conducting deep neural network learning on the abnormal state data around the base area in the step S1 to predict potential intrusion events, it includes: After performing noise elimination and reliable data screening on the flight dynamic path data and flight action posture data included in the abnormal state data, input the flight dynamic path data and the flight action posture data into the deep neural network model for analysis and learning to obtain the flight trend of the suspected foreign objects relative to the base area; According to the flight trend, predict the spatial distribution of potential intrusion events in the base area.

4. The deep learning-based drone emergency defense control method according to claim 3, wherein: When setting the inspection strategy for the drone cluster around the base area in the step S1, it includes: According to the spatial distribution of potential intrusion events in the base area and the geographical terrain distribution around the base area, identify the intruders around the base area and the spatial distribution of their intrusion locations; Set the inspection strategy for the UAV cluster around the base area according to the spatial distribution of the intrusion locations and the readiness status of the UAV clusters within the base area; wherein, the readiness status of the UAV clusters includes the parking positions and takeoff acceleration characteristics of all UAV clusters within the base area; the inspection strategy for the UAV clusters includes the inspection flight coverage spatial range of the UAV clusters around the base area.

5. The deep learning-based UAV emergency defense control method according to claim 1, wherein: In the step S2, monitor and identify the interference received locally by the UAV cluster during the inspection process, including: Listen to the external signals received locally by all UAVs under the UAV cluster during the inspection process, and extract the interference signal components of the external signals; Monitor the flight trajectory and attitude drift change data of all UAVs under the UAV cluster during the inspection process, analyze the flight trajectory and attitude drift change data, and obtain the occurrence time of the UAV flight anomaly; Compare the interference signal components in the time series of the external signals with the occurrence time of the UAV flight anomaly to identify the UAVs in the UAV cluster that have experienced flight instability events.

6. The deep learning-based UAV emergency defense control method according to claim 5, wherein: In the step S2, adjust the flight layout of the UAV cluster and the control strategy for at least some of the UAVs, including: According to the distribution position of the UAVs that have experienced flight instability events within the UAV cluster and the relative flight speed between the UAVs that have experienced flight instability events and other UAVs, adjust the flight layout of the UAV cluster, switch the UAVs that have experienced flight instability events to the internal area of the UAV cluster, and switch other UAVs to the peripheral area of the UAV cluster; According to the communication control strategy between the UAVs that have experienced flight instability events and the ground base station, switch the UAVs that have experienced flight instability events to the working state controlled by other UAVs, so that other UAVs perform positioning and flight control on the UAVs that have experienced flight instability events.

7. The deep learning-based UAV emergency defense control method according to claim 1, wherein: In the step S3, it further includes: Obtain the predicted flight trajectory of the intruders around the base area and the planned flight trajectory of the entire UAV cluster; Perform a spatial comparison between the predicted flight trajectory and the planned flight trajectory to obtain the dynamic data of the relative spatial relationship between the intruders around the base area and the UAV cluster; wherein, the dynamic data of the relative spatial relationship includes the relative spatial distance and the dynamic data of the relative spatial orientation between the intruders around the base area and the UAV cluster.

8. The deep learning-based UAV emergency defense control method according to claim 7, wherein: In the step S3, according to the dynamic data of the relative spatial relationship between the intruders around the base area and the UAV cluster, the flight states of the corresponding UAVs in the UAV cluster are adjusted, so that the intruders and the UAV cluster meet the preset position relationship conditions, including: Performing a time-evolution analysis on the dynamic data of the relative spatial distance and relative spatial orientation between the intruders around the base area and the UAV cluster, and predicting the encounter position range between the UAV cluster and the intruders; According to the encounter position range, adjusting the flight trajectories and / or flight speeds of the UAVs in the outer area of the UAV cluster, so that the UAV cluster can form an encirclement of the intruders.

9. The UAV emergency defense control method based on deep learning according to claim 1, wherein: In the step S4, according to the flight trajectory of the intruder, controlling at least one UAV in the UAV cluster to initiate non-contact interference on the intruder until the flight state of the intruder changes, including: Determining the UAV in the UAV cluster that is closest to the intruder according to the flight trajectory of the intruder; According to the relative azimuth change characteristics between the closest UAV and the intruder, controlling the operation intensity and operation duration of the closest UAV to initiate electromagnetic interference and / or laser interference on the intruder until the intruder flies away from the base area or stops flying.

10. The UAV emergency defense control system for deep learning, characterized in that, Including: A deep learning and inspection strategy setting module, configured to perform deep neural network learning on the abnormal state data around the base area, predict potential intrusion events, and thus set the inspection strategy for the UAV cluster around the base area; An interference monitoring and identification module, configured to monitor and identify the interference received locally by the UAV cluster during the inspection process; A flight layout and control strategy adjustment module, configured to adjust the flight layout of the UAV cluster and the control strategy for at least some of the UAVs; A cluster flight state adjustment module, configured to adjust the flight states of the corresponding UAVs in the UAV cluster according to the dynamic data of the relative spatial relationship between the intruders around the base area and the UAV cluster, so that the intruders and the UAV cluster meet the preset position relationship conditions; An interference initiation control module, configured to control at least one UAV in the UAV cluster to initiate non-contact interference on the intruder according to the flight trajectory of the intruder until the flight state of the intruder changes.

Images