Research On Intelligent Detection And Countermeasure Technology Of Rotor UAV Formation

Rotor unmanned aerial vehicle(UAV)formation can greatly increase its combat capability through the complementary capabilities of different units,and because of its low cost and high costeffectiveness ratio,it will become a new combat force in the future battlefield.Therefore,it is imperative to study the countermeasure technology of UAV formation.This thesis studies intelligent detection and countermeasure technology of rotor UAV formation.Because the traditional airspace detection technology can not effectively detect the formation of "low,slow,and small" UAVs,an intelligent detection method of UAV formation based on YOLOv5 and PRI transform is proposed by combining the traditional radar signal processing with deep learning.Furthermore,aiming at the lack of robustness of the reinforcement learning strategy of the conventional UAV formation agent in UAV formation confrontation,this thesis proposes an improved algorithm based on Minimax and self-game to improve the robustness of the strategy.The main research work of this dissertation includes:1.In order to solve the problem of UAV formation detection,an intelligent UAV formation detection method based on YOLOv5 and PRI transform is proposed by combining traditional radar signal processing with deep learning.Firstly,the micro-Doppler radar echo signals of the rotor UAV formation are modeled and parameterized,and then the time-frequency analysis is carried out.After that,the micro-Doppler time-frequency diagrams of the echo signals are used to extract the position information features of the scintillation pulse by the YOLOv5 network.A series of simulation experiments verify the effectiveness of the feature extraction method.Then,the position information of the scintillation pulse is mapped to the time dimension through the mapping relationship,and the arrival time of each scintillation pulse is obtained.Furthermore,the principles of the cumulative difference histogram method,the sequential difference histogram method,and the improved PRI transform method are studied.Considering that there will be missed detection or false detection in the process of extracting the scintillation pulses,the improved PRI transform algorithm with strong harmonic suppression ability is used to sort the scintillation pulses,and the number of UAVs in formation is detected.Simulation results show that,compared with the traditional feature extraction method based on CVD,the average detection performance is improved by 10.6% and the detection accuracy is more than 90% when the SNR is 20 d B and the number of UAVs in formation is less than nine,which verifies the effectiveness and application boundary of the proposed method.2.For the scenario of rotor UAV formation confrontation,since the opponent’s action strategy may change at any time,to enhance the strategy robustness of the UAV agents training model,this thesis proposes a MMSG-MADDPG algorithm based on Minimax and self-game and designs the corresponding reward and punishment function.The algorithm first trains a strong opponent for the red UAV agents through the self-game between the blue UAV agents and its "clone",so that the red UAV agents can synchronously improve their confrontation ability.At the same time,according to the Minimax idea,each training is based on the assumption that the opponent can make a perfect policy.Simulation results show that,compared with the conventional MADDPG algorithm,the performance of the proposed method is improved by about 15%,and it has stronger policy robustness.