UAV Detection And Tracking Based On Simulation Training Data

The fixed wing UAV has a wide range of applications.It has the advantages of long endurance,strong viability,large cruising area,and high maneuverability,which enable it to handle emergencies during flight.It is an important equipment for information center warfare and network security warfare,and plays a vital role in military,civil fields,such as modern warfare and aerial photography rescue,and visual inspection tasks.However,due to the limitations of various parties,there are challenges in data acquisition of fixed wing UAVs.Currently,mainstream UAV detection algorithms are rarely used to detect real fixed wing UAV images in military reconnaissance work:(1)Fixed wing UAVs are not widely used at present,so obtaining enough real data requires a lot of economic and human resources,which improves the complexity of real fixed wing UAV data acquisition.(2)The fixed wing UAV samples face various complex background environments,different angles,and different sizes in the image,which brings some noise interference and increases the complexity of target detection and tracking.In order to solve the above problems,this paper constructs a fixed wing UAV data set with the fixed wing UAV target samples as the research object.Aiming at the characteristics of the fixed wing UAV with complex background and different angle sizes,a fixed wing UAV target detection method based on simulation data migration learning is proposed.On this basis,a video target tracking learning strategy for fixed wing UAV is proposed.The main research work and achievements are as follows:1.The data set of fixed wing UAV is constructed,which includes image and video.Among them,fixed wing uav images include a large number of simulated fixed wing uav image samples and a small number of real fixed wing uav image samples.3d digital models and 2d image models are selected as simulation data.The video of fixed wing UAV includes a small number of video samples of simulated fixed wing UAV and a small number of video samples of real fixed wing UAV,and the simulation data uses a 3D digital model.2.A method of detecting real data based on weight migration of simulation training data is proposed.This method realizes target detection of real fixed wing UAV image samples by training a large number of migration weights of simulated fixed wing UAV image samples.At the same time,a two-stage target detection learning strategy is proposed.In the first stage,the network model uses a single-stage end-to-end detection algorithm to extract the high-level features of image samples.In the second stage,the network model further detects targets of different scales through a multi-layer feature pyramid.This strategy achieves a low false detection rate,and achieves high accuracy,recall,average accuracy,and the number of frames transmitted per second.3.The video target tracking learning strategy of fixed wing UAV is proposed to realize the action recognition of simulation and real fixed wing UAV video.Firstly,the strategy combines target detection and target tracking to obtain real-time position and accurate scale information of target fixed wing UAV;After obtaining the location and scale information,the small sample learning method is used to accurately identify the target’s action on the time series and carry out semantic coding.This strategy realizes the better recognition of the sample action sequence in the simulation and real fixed wing UAV video,and obtains higher accuracy.In conclusion,this preliminary work has potential application prospects in the research of target detection and target tracking of fixed wing UAV.